Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <linux/utsname.h>
17#include <linux/stacktrace.h>
18#include <linux/writeback.h>
19#include <linux/kallsyms.h>
20#include <linux/security.h>
21#include <linux/seq_file.h>
22#include <linux/irqflags.h>
23#include <linux/debugfs.h>
24#include <linux/tracefs.h>
25#include <linux/pagemap.h>
26#include <linux/hardirq.h>
27#include <linux/linkage.h>
28#include <linux/uaccess.h>
29#include <linux/cleanup.h>
30#include <linux/vmalloc.h>
31#include <linux/ftrace.h>
32#include <linux/module.h>
33#include <linux/percpu.h>
34#include <linux/splice.h>
35#include <linux/kdebug.h>
36#include <linux/string.h>
37#include <linux/mount.h>
38#include <linux/rwsem.h>
39#include <linux/slab.h>
40#include <linux/ctype.h>
41#include <linux/init.h>
42#include <linux/panic_notifier.h>
43#include <linux/poll.h>
44#include <linux/nmi.h>
45#include <linux/fs.h>
46#include <linux/trace.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/rt.h>
49#include <linux/fsnotify.h>
50#include <linux/irq_work.h>
51#include <linux/workqueue.h>
52
53#include <asm/setup.h> /* COMMAND_LINE_SIZE */
54
55#include "trace.h"
56#include "trace_output.h"
57
58#ifdef CONFIG_FTRACE_STARTUP_TEST
59/*
60 * We need to change this state when a selftest is running.
61 * A selftest will lurk into the ring-buffer to count the
62 * entries inserted during the selftest although some concurrent
63 * insertions into the ring-buffer such as trace_printk could occurred
64 * at the same time, giving false positive or negative results.
65 */
66static bool __read_mostly tracing_selftest_running;
67
68/*
69 * If boot-time tracing including tracers/events via kernel cmdline
70 * is running, we do not want to run SELFTEST.
71 */
72bool __read_mostly tracing_selftest_disabled;
73
74void __init disable_tracing_selftest(const char *reason)
75{
76 if (!tracing_selftest_disabled) {
77 tracing_selftest_disabled = true;
78 pr_info("Ftrace startup test is disabled due to %s\n", reason);
79 }
80}
81#else
82#define tracing_selftest_running 0
83#define tracing_selftest_disabled 0
84#endif
85
86/* Pipe tracepoints to printk */
87static struct trace_iterator *tracepoint_print_iter;
88int tracepoint_printk;
89static bool tracepoint_printk_stop_on_boot __initdata;
90static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91
92/* For tracers that don't implement custom flags */
93static struct tracer_opt dummy_tracer_opt[] = {
94 { }
95};
96
97static int
98dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
99{
100 return 0;
101}
102
103/*
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
106 * occurred.
107 */
108DEFINE_PER_CPU(bool, trace_taskinfo_save);
109
110/*
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
114 * this back to zero.
115 */
116static int tracing_disabled = 1;
117
118cpumask_var_t __read_mostly tracing_buffer_mask;
119
120/*
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 *
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
127 * serial console.
128 *
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
134 * Set instance name if you want to dump the specific trace instance
135 * Multiple instance dump is also supported, and instances are seperated
136 * by commas.
137 */
138/* Set to string format zero to disable by default */
139char ftrace_dump_on_oops[MAX_TRACER_SIZE] = "0";
140
141/* When set, tracing will stop when a WARN*() is hit */
142int __disable_trace_on_warning;
143
144#ifdef CONFIG_TRACE_EVAL_MAP_FILE
145/* Map of enums to their values, for "eval_map" file */
146struct trace_eval_map_head {
147 struct module *mod;
148 unsigned long length;
149};
150
151union trace_eval_map_item;
152
153struct trace_eval_map_tail {
154 /*
155 * "end" is first and points to NULL as it must be different
156 * than "mod" or "eval_string"
157 */
158 union trace_eval_map_item *next;
159 const char *end; /* points to NULL */
160};
161
162static DEFINE_MUTEX(trace_eval_mutex);
163
164/*
165 * The trace_eval_maps are saved in an array with two extra elements,
166 * one at the beginning, and one at the end. The beginning item contains
167 * the count of the saved maps (head.length), and the module they
168 * belong to if not built in (head.mod). The ending item contains a
169 * pointer to the next array of saved eval_map items.
170 */
171union trace_eval_map_item {
172 struct trace_eval_map map;
173 struct trace_eval_map_head head;
174 struct trace_eval_map_tail tail;
175};
176
177static union trace_eval_map_item *trace_eval_maps;
178#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
179
180int tracing_set_tracer(struct trace_array *tr, const char *buf);
181static void ftrace_trace_userstack(struct trace_array *tr,
182 struct trace_buffer *buffer,
183 unsigned int trace_ctx);
184
185static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
186static char *default_bootup_tracer;
187
188static bool allocate_snapshot;
189static bool snapshot_at_boot;
190
191static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
192static int boot_instance_index;
193
194static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
195static int boot_snapshot_index;
196
197static int __init set_cmdline_ftrace(char *str)
198{
199 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
200 default_bootup_tracer = bootup_tracer_buf;
201 /* We are using ftrace early, expand it */
202 trace_set_ring_buffer_expanded(NULL);
203 return 1;
204}
205__setup("ftrace=", set_cmdline_ftrace);
206
207int ftrace_dump_on_oops_enabled(void)
208{
209 if (!strcmp("0", ftrace_dump_on_oops))
210 return 0;
211 else
212 return 1;
213}
214
215static int __init set_ftrace_dump_on_oops(char *str)
216{
217 if (!*str) {
218 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
219 return 1;
220 }
221
222 if (*str == ',') {
223 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
224 strscpy(ftrace_dump_on_oops + 1, str, MAX_TRACER_SIZE - 1);
225 return 1;
226 }
227
228 if (*str++ == '=') {
229 strscpy(ftrace_dump_on_oops, str, MAX_TRACER_SIZE);
230 return 1;
231 }
232
233 return 0;
234}
235__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
236
237static int __init stop_trace_on_warning(char *str)
238{
239 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
240 __disable_trace_on_warning = 1;
241 return 1;
242}
243__setup("traceoff_on_warning", stop_trace_on_warning);
244
245static int __init boot_alloc_snapshot(char *str)
246{
247 char *slot = boot_snapshot_info + boot_snapshot_index;
248 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
249 int ret;
250
251 if (str[0] == '=') {
252 str++;
253 if (strlen(str) >= left)
254 return -1;
255
256 ret = snprintf(slot, left, "%s\t", str);
257 boot_snapshot_index += ret;
258 } else {
259 allocate_snapshot = true;
260 /* We also need the main ring buffer expanded */
261 trace_set_ring_buffer_expanded(NULL);
262 }
263 return 1;
264}
265__setup("alloc_snapshot", boot_alloc_snapshot);
266
267
268static int __init boot_snapshot(char *str)
269{
270 snapshot_at_boot = true;
271 boot_alloc_snapshot(str);
272 return 1;
273}
274__setup("ftrace_boot_snapshot", boot_snapshot);
275
276
277static int __init boot_instance(char *str)
278{
279 char *slot = boot_instance_info + boot_instance_index;
280 int left = sizeof(boot_instance_info) - boot_instance_index;
281 int ret;
282
283 if (strlen(str) >= left)
284 return -1;
285
286 ret = snprintf(slot, left, "%s\t", str);
287 boot_instance_index += ret;
288
289 return 1;
290}
291__setup("trace_instance=", boot_instance);
292
293
294static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
295
296static int __init set_trace_boot_options(char *str)
297{
298 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
299 return 1;
300}
301__setup("trace_options=", set_trace_boot_options);
302
303static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
304static char *trace_boot_clock __initdata;
305
306static int __init set_trace_boot_clock(char *str)
307{
308 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
309 trace_boot_clock = trace_boot_clock_buf;
310 return 1;
311}
312__setup("trace_clock=", set_trace_boot_clock);
313
314static int __init set_tracepoint_printk(char *str)
315{
316 /* Ignore the "tp_printk_stop_on_boot" param */
317 if (*str == '_')
318 return 0;
319
320 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
321 tracepoint_printk = 1;
322 return 1;
323}
324__setup("tp_printk", set_tracepoint_printk);
325
326static int __init set_tracepoint_printk_stop(char *str)
327{
328 tracepoint_printk_stop_on_boot = true;
329 return 1;
330}
331__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
332
333unsigned long long ns2usecs(u64 nsec)
334{
335 nsec += 500;
336 do_div(nsec, 1000);
337 return nsec;
338}
339
340static void
341trace_process_export(struct trace_export *export,
342 struct ring_buffer_event *event, int flag)
343{
344 struct trace_entry *entry;
345 unsigned int size = 0;
346
347 if (export->flags & flag) {
348 entry = ring_buffer_event_data(event);
349 size = ring_buffer_event_length(event);
350 export->write(export, entry, size);
351 }
352}
353
354static DEFINE_MUTEX(ftrace_export_lock);
355
356static struct trace_export __rcu *ftrace_exports_list __read_mostly;
357
358static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
359static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
360static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
361
362static inline void ftrace_exports_enable(struct trace_export *export)
363{
364 if (export->flags & TRACE_EXPORT_FUNCTION)
365 static_branch_inc(&trace_function_exports_enabled);
366
367 if (export->flags & TRACE_EXPORT_EVENT)
368 static_branch_inc(&trace_event_exports_enabled);
369
370 if (export->flags & TRACE_EXPORT_MARKER)
371 static_branch_inc(&trace_marker_exports_enabled);
372}
373
374static inline void ftrace_exports_disable(struct trace_export *export)
375{
376 if (export->flags & TRACE_EXPORT_FUNCTION)
377 static_branch_dec(&trace_function_exports_enabled);
378
379 if (export->flags & TRACE_EXPORT_EVENT)
380 static_branch_dec(&trace_event_exports_enabled);
381
382 if (export->flags & TRACE_EXPORT_MARKER)
383 static_branch_dec(&trace_marker_exports_enabled);
384}
385
386static void ftrace_exports(struct ring_buffer_event *event, int flag)
387{
388 struct trace_export *export;
389
390 preempt_disable_notrace();
391
392 export = rcu_dereference_raw_check(ftrace_exports_list);
393 while (export) {
394 trace_process_export(export, event, flag);
395 export = rcu_dereference_raw_check(export->next);
396 }
397
398 preempt_enable_notrace();
399}
400
401static inline void
402add_trace_export(struct trace_export **list, struct trace_export *export)
403{
404 rcu_assign_pointer(export->next, *list);
405 /*
406 * We are entering export into the list but another
407 * CPU might be walking that list. We need to make sure
408 * the export->next pointer is valid before another CPU sees
409 * the export pointer included into the list.
410 */
411 rcu_assign_pointer(*list, export);
412}
413
414static inline int
415rm_trace_export(struct trace_export **list, struct trace_export *export)
416{
417 struct trace_export **p;
418
419 for (p = list; *p != NULL; p = &(*p)->next)
420 if (*p == export)
421 break;
422
423 if (*p != export)
424 return -1;
425
426 rcu_assign_pointer(*p, (*p)->next);
427
428 return 0;
429}
430
431static inline void
432add_ftrace_export(struct trace_export **list, struct trace_export *export)
433{
434 ftrace_exports_enable(export);
435
436 add_trace_export(list, export);
437}
438
439static inline int
440rm_ftrace_export(struct trace_export **list, struct trace_export *export)
441{
442 int ret;
443
444 ret = rm_trace_export(list, export);
445 ftrace_exports_disable(export);
446
447 return ret;
448}
449
450int register_ftrace_export(struct trace_export *export)
451{
452 if (WARN_ON_ONCE(!export->write))
453 return -1;
454
455 mutex_lock(&ftrace_export_lock);
456
457 add_ftrace_export(&ftrace_exports_list, export);
458
459 mutex_unlock(&ftrace_export_lock);
460
461 return 0;
462}
463EXPORT_SYMBOL_GPL(register_ftrace_export);
464
465int unregister_ftrace_export(struct trace_export *export)
466{
467 int ret;
468
469 mutex_lock(&ftrace_export_lock);
470
471 ret = rm_ftrace_export(&ftrace_exports_list, export);
472
473 mutex_unlock(&ftrace_export_lock);
474
475 return ret;
476}
477EXPORT_SYMBOL_GPL(unregister_ftrace_export);
478
479/* trace_flags holds trace_options default values */
480#define TRACE_DEFAULT_FLAGS \
481 (FUNCTION_DEFAULT_FLAGS | \
482 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
483 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
484 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
485 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
486 TRACE_ITER_HASH_PTR | TRACE_ITER_TRACE_PRINTK)
487
488/* trace_options that are only supported by global_trace */
489#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
490 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
491
492/* trace_flags that are default zero for instances */
493#define ZEROED_TRACE_FLAGS \
494 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK | TRACE_ITER_TRACE_PRINTK)
495
496/*
497 * The global_trace is the descriptor that holds the top-level tracing
498 * buffers for the live tracing.
499 */
500static struct trace_array global_trace = {
501 .trace_flags = TRACE_DEFAULT_FLAGS,
502};
503
504static struct trace_array *printk_trace = &global_trace;
505
506static __always_inline bool printk_binsafe(struct trace_array *tr)
507{
508 /*
509 * The binary format of traceprintk can cause a crash if used
510 * by a buffer from another boot. Force the use of the
511 * non binary version of trace_printk if the trace_printk
512 * buffer is a boot mapped ring buffer.
513 */
514 return !(tr->flags & TRACE_ARRAY_FL_BOOT);
515}
516
517static void update_printk_trace(struct trace_array *tr)
518{
519 if (printk_trace == tr)
520 return;
521
522 printk_trace->trace_flags &= ~TRACE_ITER_TRACE_PRINTK;
523 printk_trace = tr;
524 tr->trace_flags |= TRACE_ITER_TRACE_PRINTK;
525}
526
527void trace_set_ring_buffer_expanded(struct trace_array *tr)
528{
529 if (!tr)
530 tr = &global_trace;
531 tr->ring_buffer_expanded = true;
532}
533
534LIST_HEAD(ftrace_trace_arrays);
535
536int trace_array_get(struct trace_array *this_tr)
537{
538 struct trace_array *tr;
539
540 guard(mutex)(&trace_types_lock);
541 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
542 if (tr == this_tr) {
543 tr->ref++;
544 return 0;
545 }
546 }
547
548 return -ENODEV;
549}
550
551static void __trace_array_put(struct trace_array *this_tr)
552{
553 WARN_ON(!this_tr->ref);
554 this_tr->ref--;
555}
556
557/**
558 * trace_array_put - Decrement the reference counter for this trace array.
559 * @this_tr : pointer to the trace array
560 *
561 * NOTE: Use this when we no longer need the trace array returned by
562 * trace_array_get_by_name(). This ensures the trace array can be later
563 * destroyed.
564 *
565 */
566void trace_array_put(struct trace_array *this_tr)
567{
568 if (!this_tr)
569 return;
570
571 mutex_lock(&trace_types_lock);
572 __trace_array_put(this_tr);
573 mutex_unlock(&trace_types_lock);
574}
575EXPORT_SYMBOL_GPL(trace_array_put);
576
577int tracing_check_open_get_tr(struct trace_array *tr)
578{
579 int ret;
580
581 ret = security_locked_down(LOCKDOWN_TRACEFS);
582 if (ret)
583 return ret;
584
585 if (tracing_disabled)
586 return -ENODEV;
587
588 if (tr && trace_array_get(tr) < 0)
589 return -ENODEV;
590
591 return 0;
592}
593
594/**
595 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
596 * @filtered_pids: The list of pids to check
597 * @search_pid: The PID to find in @filtered_pids
598 *
599 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
600 */
601bool
602trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
603{
604 return trace_pid_list_is_set(filtered_pids, search_pid);
605}
606
607/**
608 * trace_ignore_this_task - should a task be ignored for tracing
609 * @filtered_pids: The list of pids to check
610 * @filtered_no_pids: The list of pids not to be traced
611 * @task: The task that should be ignored if not filtered
612 *
613 * Checks if @task should be traced or not from @filtered_pids.
614 * Returns true if @task should *NOT* be traced.
615 * Returns false if @task should be traced.
616 */
617bool
618trace_ignore_this_task(struct trace_pid_list *filtered_pids,
619 struct trace_pid_list *filtered_no_pids,
620 struct task_struct *task)
621{
622 /*
623 * If filtered_no_pids is not empty, and the task's pid is listed
624 * in filtered_no_pids, then return true.
625 * Otherwise, if filtered_pids is empty, that means we can
626 * trace all tasks. If it has content, then only trace pids
627 * within filtered_pids.
628 */
629
630 return (filtered_pids &&
631 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
632 (filtered_no_pids &&
633 trace_find_filtered_pid(filtered_no_pids, task->pid));
634}
635
636/**
637 * trace_filter_add_remove_task - Add or remove a task from a pid_list
638 * @pid_list: The list to modify
639 * @self: The current task for fork or NULL for exit
640 * @task: The task to add or remove
641 *
642 * If adding a task, if @self is defined, the task is only added if @self
643 * is also included in @pid_list. This happens on fork and tasks should
644 * only be added when the parent is listed. If @self is NULL, then the
645 * @task pid will be removed from the list, which would happen on exit
646 * of a task.
647 */
648void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
649 struct task_struct *self,
650 struct task_struct *task)
651{
652 if (!pid_list)
653 return;
654
655 /* For forks, we only add if the forking task is listed */
656 if (self) {
657 if (!trace_find_filtered_pid(pid_list, self->pid))
658 return;
659 }
660
661 /* "self" is set for forks, and NULL for exits */
662 if (self)
663 trace_pid_list_set(pid_list, task->pid);
664 else
665 trace_pid_list_clear(pid_list, task->pid);
666}
667
668/**
669 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
670 * @pid_list: The pid list to show
671 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
672 * @pos: The position of the file
673 *
674 * This is used by the seq_file "next" operation to iterate the pids
675 * listed in a trace_pid_list structure.
676 *
677 * Returns the pid+1 as we want to display pid of zero, but NULL would
678 * stop the iteration.
679 */
680void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
681{
682 long pid = (unsigned long)v;
683 unsigned int next;
684
685 (*pos)++;
686
687 /* pid already is +1 of the actual previous bit */
688 if (trace_pid_list_next(pid_list, pid, &next) < 0)
689 return NULL;
690
691 pid = next;
692
693 /* Return pid + 1 to allow zero to be represented */
694 return (void *)(pid + 1);
695}
696
697/**
698 * trace_pid_start - Used for seq_file to start reading pid lists
699 * @pid_list: The pid list to show
700 * @pos: The position of the file
701 *
702 * This is used by seq_file "start" operation to start the iteration
703 * of listing pids.
704 *
705 * Returns the pid+1 as we want to display pid of zero, but NULL would
706 * stop the iteration.
707 */
708void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
709{
710 unsigned long pid;
711 unsigned int first;
712 loff_t l = 0;
713
714 if (trace_pid_list_first(pid_list, &first) < 0)
715 return NULL;
716
717 pid = first;
718
719 /* Return pid + 1 so that zero can be the exit value */
720 for (pid++; pid && l < *pos;
721 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
722 ;
723 return (void *)pid;
724}
725
726/**
727 * trace_pid_show - show the current pid in seq_file processing
728 * @m: The seq_file structure to write into
729 * @v: A void pointer of the pid (+1) value to display
730 *
731 * Can be directly used by seq_file operations to display the current
732 * pid value.
733 */
734int trace_pid_show(struct seq_file *m, void *v)
735{
736 unsigned long pid = (unsigned long)v - 1;
737
738 seq_printf(m, "%lu\n", pid);
739 return 0;
740}
741
742/* 128 should be much more than enough */
743#define PID_BUF_SIZE 127
744
745int trace_pid_write(struct trace_pid_list *filtered_pids,
746 struct trace_pid_list **new_pid_list,
747 const char __user *ubuf, size_t cnt)
748{
749 struct trace_pid_list *pid_list;
750 struct trace_parser parser;
751 unsigned long val;
752 int nr_pids = 0;
753 ssize_t read = 0;
754 ssize_t ret;
755 loff_t pos;
756 pid_t pid;
757
758 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
759 return -ENOMEM;
760
761 /*
762 * Always recreate a new array. The write is an all or nothing
763 * operation. Always create a new array when adding new pids by
764 * the user. If the operation fails, then the current list is
765 * not modified.
766 */
767 pid_list = trace_pid_list_alloc();
768 if (!pid_list) {
769 trace_parser_put(&parser);
770 return -ENOMEM;
771 }
772
773 if (filtered_pids) {
774 /* copy the current bits to the new max */
775 ret = trace_pid_list_first(filtered_pids, &pid);
776 while (!ret) {
777 trace_pid_list_set(pid_list, pid);
778 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
779 nr_pids++;
780 }
781 }
782
783 ret = 0;
784 while (cnt > 0) {
785
786 pos = 0;
787
788 ret = trace_get_user(&parser, ubuf, cnt, &pos);
789 if (ret < 0)
790 break;
791
792 read += ret;
793 ubuf += ret;
794 cnt -= ret;
795
796 if (!trace_parser_loaded(&parser))
797 break;
798
799 ret = -EINVAL;
800 if (kstrtoul(parser.buffer, 0, &val))
801 break;
802
803 pid = (pid_t)val;
804
805 if (trace_pid_list_set(pid_list, pid) < 0) {
806 ret = -1;
807 break;
808 }
809 nr_pids++;
810
811 trace_parser_clear(&parser);
812 ret = 0;
813 }
814 trace_parser_put(&parser);
815
816 if (ret < 0) {
817 trace_pid_list_free(pid_list);
818 return ret;
819 }
820
821 if (!nr_pids) {
822 /* Cleared the list of pids */
823 trace_pid_list_free(pid_list);
824 pid_list = NULL;
825 }
826
827 *new_pid_list = pid_list;
828
829 return read;
830}
831
832static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
833{
834 u64 ts;
835
836 /* Early boot up does not have a buffer yet */
837 if (!buf->buffer)
838 return trace_clock_local();
839
840 ts = ring_buffer_time_stamp(buf->buffer);
841 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
842
843 return ts;
844}
845
846u64 ftrace_now(int cpu)
847{
848 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
849}
850
851/**
852 * tracing_is_enabled - Show if global_trace has been enabled
853 *
854 * Shows if the global trace has been enabled or not. It uses the
855 * mirror flag "buffer_disabled" to be used in fast paths such as for
856 * the irqsoff tracer. But it may be inaccurate due to races. If you
857 * need to know the accurate state, use tracing_is_on() which is a little
858 * slower, but accurate.
859 */
860int tracing_is_enabled(void)
861{
862 /*
863 * For quick access (irqsoff uses this in fast path), just
864 * return the mirror variable of the state of the ring buffer.
865 * It's a little racy, but we don't really care.
866 */
867 smp_rmb();
868 return !global_trace.buffer_disabled;
869}
870
871/*
872 * trace_buf_size is the size in bytes that is allocated
873 * for a buffer. Note, the number of bytes is always rounded
874 * to page size.
875 *
876 * This number is purposely set to a low number of 16384.
877 * If the dump on oops happens, it will be much appreciated
878 * to not have to wait for all that output. Anyway this can be
879 * boot time and run time configurable.
880 */
881#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
882
883static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
884
885/* trace_types holds a link list of available tracers. */
886static struct tracer *trace_types __read_mostly;
887
888/*
889 * trace_types_lock is used to protect the trace_types list.
890 */
891DEFINE_MUTEX(trace_types_lock);
892
893/*
894 * serialize the access of the ring buffer
895 *
896 * ring buffer serializes readers, but it is low level protection.
897 * The validity of the events (which returns by ring_buffer_peek() ..etc)
898 * are not protected by ring buffer.
899 *
900 * The content of events may become garbage if we allow other process consumes
901 * these events concurrently:
902 * A) the page of the consumed events may become a normal page
903 * (not reader page) in ring buffer, and this page will be rewritten
904 * by events producer.
905 * B) The page of the consumed events may become a page for splice_read,
906 * and this page will be returned to system.
907 *
908 * These primitives allow multi process access to different cpu ring buffer
909 * concurrently.
910 *
911 * These primitives don't distinguish read-only and read-consume access.
912 * Multi read-only access are also serialized.
913 */
914
915#ifdef CONFIG_SMP
916static DECLARE_RWSEM(all_cpu_access_lock);
917static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
918
919static inline void trace_access_lock(int cpu)
920{
921 if (cpu == RING_BUFFER_ALL_CPUS) {
922 /* gain it for accessing the whole ring buffer. */
923 down_write(&all_cpu_access_lock);
924 } else {
925 /* gain it for accessing a cpu ring buffer. */
926
927 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
928 down_read(&all_cpu_access_lock);
929
930 /* Secondly block other access to this @cpu ring buffer. */
931 mutex_lock(&per_cpu(cpu_access_lock, cpu));
932 }
933}
934
935static inline void trace_access_unlock(int cpu)
936{
937 if (cpu == RING_BUFFER_ALL_CPUS) {
938 up_write(&all_cpu_access_lock);
939 } else {
940 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
941 up_read(&all_cpu_access_lock);
942 }
943}
944
945static inline void trace_access_lock_init(void)
946{
947 int cpu;
948
949 for_each_possible_cpu(cpu)
950 mutex_init(&per_cpu(cpu_access_lock, cpu));
951}
952
953#else
954
955static DEFINE_MUTEX(access_lock);
956
957static inline void trace_access_lock(int cpu)
958{
959 (void)cpu;
960 mutex_lock(&access_lock);
961}
962
963static inline void trace_access_unlock(int cpu)
964{
965 (void)cpu;
966 mutex_unlock(&access_lock);
967}
968
969static inline void trace_access_lock_init(void)
970{
971}
972
973#endif
974
975#ifdef CONFIG_STACKTRACE
976static void __ftrace_trace_stack(struct trace_array *tr,
977 struct trace_buffer *buffer,
978 unsigned int trace_ctx,
979 int skip, struct pt_regs *regs);
980static inline void ftrace_trace_stack(struct trace_array *tr,
981 struct trace_buffer *buffer,
982 unsigned int trace_ctx,
983 int skip, struct pt_regs *regs);
984
985#else
986static inline void __ftrace_trace_stack(struct trace_array *tr,
987 struct trace_buffer *buffer,
988 unsigned int trace_ctx,
989 int skip, struct pt_regs *regs)
990{
991}
992static inline void ftrace_trace_stack(struct trace_array *tr,
993 struct trace_buffer *buffer,
994 unsigned long trace_ctx,
995 int skip, struct pt_regs *regs)
996{
997}
998
999#endif
1000
1001static __always_inline void
1002trace_event_setup(struct ring_buffer_event *event,
1003 int type, unsigned int trace_ctx)
1004{
1005 struct trace_entry *ent = ring_buffer_event_data(event);
1006
1007 tracing_generic_entry_update(ent, type, trace_ctx);
1008}
1009
1010static __always_inline struct ring_buffer_event *
1011__trace_buffer_lock_reserve(struct trace_buffer *buffer,
1012 int type,
1013 unsigned long len,
1014 unsigned int trace_ctx)
1015{
1016 struct ring_buffer_event *event;
1017
1018 event = ring_buffer_lock_reserve(buffer, len);
1019 if (event != NULL)
1020 trace_event_setup(event, type, trace_ctx);
1021
1022 return event;
1023}
1024
1025void tracer_tracing_on(struct trace_array *tr)
1026{
1027 if (tr->array_buffer.buffer)
1028 ring_buffer_record_on(tr->array_buffer.buffer);
1029 /*
1030 * This flag is looked at when buffers haven't been allocated
1031 * yet, or by some tracers (like irqsoff), that just want to
1032 * know if the ring buffer has been disabled, but it can handle
1033 * races of where it gets disabled but we still do a record.
1034 * As the check is in the fast path of the tracers, it is more
1035 * important to be fast than accurate.
1036 */
1037 tr->buffer_disabled = 0;
1038 /* Make the flag seen by readers */
1039 smp_wmb();
1040}
1041
1042/**
1043 * tracing_on - enable tracing buffers
1044 *
1045 * This function enables tracing buffers that may have been
1046 * disabled with tracing_off.
1047 */
1048void tracing_on(void)
1049{
1050 tracer_tracing_on(&global_trace);
1051}
1052EXPORT_SYMBOL_GPL(tracing_on);
1053
1054
1055static __always_inline void
1056__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1057{
1058 __this_cpu_write(trace_taskinfo_save, true);
1059
1060 /* If this is the temp buffer, we need to commit fully */
1061 if (this_cpu_read(trace_buffered_event) == event) {
1062 /* Length is in event->array[0] */
1063 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1064 /* Release the temp buffer */
1065 this_cpu_dec(trace_buffered_event_cnt);
1066 /* ring_buffer_unlock_commit() enables preemption */
1067 preempt_enable_notrace();
1068 } else
1069 ring_buffer_unlock_commit(buffer);
1070}
1071
1072int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1073 const char *str, int size)
1074{
1075 struct ring_buffer_event *event;
1076 struct trace_buffer *buffer;
1077 struct print_entry *entry;
1078 unsigned int trace_ctx;
1079 int alloc;
1080
1081 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1082 return 0;
1083
1084 if (unlikely(tracing_selftest_running && tr == &global_trace))
1085 return 0;
1086
1087 if (unlikely(tracing_disabled))
1088 return 0;
1089
1090 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1091
1092 trace_ctx = tracing_gen_ctx();
1093 buffer = tr->array_buffer.buffer;
1094 ring_buffer_nest_start(buffer);
1095 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1096 trace_ctx);
1097 if (!event) {
1098 size = 0;
1099 goto out;
1100 }
1101
1102 entry = ring_buffer_event_data(event);
1103 entry->ip = ip;
1104
1105 memcpy(&entry->buf, str, size);
1106
1107 /* Add a newline if necessary */
1108 if (entry->buf[size - 1] != '\n') {
1109 entry->buf[size] = '\n';
1110 entry->buf[size + 1] = '\0';
1111 } else
1112 entry->buf[size] = '\0';
1113
1114 __buffer_unlock_commit(buffer, event);
1115 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1116 out:
1117 ring_buffer_nest_end(buffer);
1118 return size;
1119}
1120EXPORT_SYMBOL_GPL(__trace_array_puts);
1121
1122/**
1123 * __trace_puts - write a constant string into the trace buffer.
1124 * @ip: The address of the caller
1125 * @str: The constant string to write
1126 * @size: The size of the string.
1127 */
1128int __trace_puts(unsigned long ip, const char *str, int size)
1129{
1130 return __trace_array_puts(printk_trace, ip, str, size);
1131}
1132EXPORT_SYMBOL_GPL(__trace_puts);
1133
1134/**
1135 * __trace_bputs - write the pointer to a constant string into trace buffer
1136 * @ip: The address of the caller
1137 * @str: The constant string to write to the buffer to
1138 */
1139int __trace_bputs(unsigned long ip, const char *str)
1140{
1141 struct trace_array *tr = READ_ONCE(printk_trace);
1142 struct ring_buffer_event *event;
1143 struct trace_buffer *buffer;
1144 struct bputs_entry *entry;
1145 unsigned int trace_ctx;
1146 int size = sizeof(struct bputs_entry);
1147 int ret = 0;
1148
1149 if (!printk_binsafe(tr))
1150 return __trace_puts(ip, str, strlen(str));
1151
1152 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1153 return 0;
1154
1155 if (unlikely(tracing_selftest_running || tracing_disabled))
1156 return 0;
1157
1158 trace_ctx = tracing_gen_ctx();
1159 buffer = tr->array_buffer.buffer;
1160
1161 ring_buffer_nest_start(buffer);
1162 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1163 trace_ctx);
1164 if (!event)
1165 goto out;
1166
1167 entry = ring_buffer_event_data(event);
1168 entry->ip = ip;
1169 entry->str = str;
1170
1171 __buffer_unlock_commit(buffer, event);
1172 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1173
1174 ret = 1;
1175 out:
1176 ring_buffer_nest_end(buffer);
1177 return ret;
1178}
1179EXPORT_SYMBOL_GPL(__trace_bputs);
1180
1181#ifdef CONFIG_TRACER_SNAPSHOT
1182static void tracing_snapshot_instance_cond(struct trace_array *tr,
1183 void *cond_data)
1184{
1185 struct tracer *tracer = tr->current_trace;
1186 unsigned long flags;
1187
1188 if (in_nmi()) {
1189 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1190 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1191 return;
1192 }
1193
1194 if (!tr->allocated_snapshot) {
1195 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1196 trace_array_puts(tr, "*** stopping trace here! ***\n");
1197 tracer_tracing_off(tr);
1198 return;
1199 }
1200
1201 /* Note, snapshot can not be used when the tracer uses it */
1202 if (tracer->use_max_tr) {
1203 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1204 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1205 return;
1206 }
1207
1208 if (tr->mapped) {
1209 trace_array_puts(tr, "*** BUFFER MEMORY MAPPED ***\n");
1210 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1211 return;
1212 }
1213
1214 local_irq_save(flags);
1215 update_max_tr(tr, current, smp_processor_id(), cond_data);
1216 local_irq_restore(flags);
1217}
1218
1219void tracing_snapshot_instance(struct trace_array *tr)
1220{
1221 tracing_snapshot_instance_cond(tr, NULL);
1222}
1223
1224/**
1225 * tracing_snapshot - take a snapshot of the current buffer.
1226 *
1227 * This causes a swap between the snapshot buffer and the current live
1228 * tracing buffer. You can use this to take snapshots of the live
1229 * trace when some condition is triggered, but continue to trace.
1230 *
1231 * Note, make sure to allocate the snapshot with either
1232 * a tracing_snapshot_alloc(), or by doing it manually
1233 * with: echo 1 > /sys/kernel/tracing/snapshot
1234 *
1235 * If the snapshot buffer is not allocated, it will stop tracing.
1236 * Basically making a permanent snapshot.
1237 */
1238void tracing_snapshot(void)
1239{
1240 struct trace_array *tr = &global_trace;
1241
1242 tracing_snapshot_instance(tr);
1243}
1244EXPORT_SYMBOL_GPL(tracing_snapshot);
1245
1246/**
1247 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1248 * @tr: The tracing instance to snapshot
1249 * @cond_data: The data to be tested conditionally, and possibly saved
1250 *
1251 * This is the same as tracing_snapshot() except that the snapshot is
1252 * conditional - the snapshot will only happen if the
1253 * cond_snapshot.update() implementation receiving the cond_data
1254 * returns true, which means that the trace array's cond_snapshot
1255 * update() operation used the cond_data to determine whether the
1256 * snapshot should be taken, and if it was, presumably saved it along
1257 * with the snapshot.
1258 */
1259void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1260{
1261 tracing_snapshot_instance_cond(tr, cond_data);
1262}
1263EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1264
1265/**
1266 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1267 * @tr: The tracing instance
1268 *
1269 * When the user enables a conditional snapshot using
1270 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1271 * with the snapshot. This accessor is used to retrieve it.
1272 *
1273 * Should not be called from cond_snapshot.update(), since it takes
1274 * the tr->max_lock lock, which the code calling
1275 * cond_snapshot.update() has already done.
1276 *
1277 * Returns the cond_data associated with the trace array's snapshot.
1278 */
1279void *tracing_cond_snapshot_data(struct trace_array *tr)
1280{
1281 void *cond_data = NULL;
1282
1283 local_irq_disable();
1284 arch_spin_lock(&tr->max_lock);
1285
1286 if (tr->cond_snapshot)
1287 cond_data = tr->cond_snapshot->cond_data;
1288
1289 arch_spin_unlock(&tr->max_lock);
1290 local_irq_enable();
1291
1292 return cond_data;
1293}
1294EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1295
1296static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1297 struct array_buffer *size_buf, int cpu_id);
1298static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1299
1300int tracing_alloc_snapshot_instance(struct trace_array *tr)
1301{
1302 int order;
1303 int ret;
1304
1305 if (!tr->allocated_snapshot) {
1306
1307 /* Make the snapshot buffer have the same order as main buffer */
1308 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1309 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1310 if (ret < 0)
1311 return ret;
1312
1313 /* allocate spare buffer */
1314 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1315 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1316 if (ret < 0)
1317 return ret;
1318
1319 tr->allocated_snapshot = true;
1320 }
1321
1322 return 0;
1323}
1324
1325static void free_snapshot(struct trace_array *tr)
1326{
1327 /*
1328 * We don't free the ring buffer. instead, resize it because
1329 * The max_tr ring buffer has some state (e.g. ring->clock) and
1330 * we want preserve it.
1331 */
1332 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1333 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1334 set_buffer_entries(&tr->max_buffer, 1);
1335 tracing_reset_online_cpus(&tr->max_buffer);
1336 tr->allocated_snapshot = false;
1337}
1338
1339static int tracing_arm_snapshot_locked(struct trace_array *tr)
1340{
1341 int ret;
1342
1343 lockdep_assert_held(&trace_types_lock);
1344
1345 spin_lock(&tr->snapshot_trigger_lock);
1346 if (tr->snapshot == UINT_MAX || tr->mapped) {
1347 spin_unlock(&tr->snapshot_trigger_lock);
1348 return -EBUSY;
1349 }
1350
1351 tr->snapshot++;
1352 spin_unlock(&tr->snapshot_trigger_lock);
1353
1354 ret = tracing_alloc_snapshot_instance(tr);
1355 if (ret) {
1356 spin_lock(&tr->snapshot_trigger_lock);
1357 tr->snapshot--;
1358 spin_unlock(&tr->snapshot_trigger_lock);
1359 }
1360
1361 return ret;
1362}
1363
1364int tracing_arm_snapshot(struct trace_array *tr)
1365{
1366 int ret;
1367
1368 mutex_lock(&trace_types_lock);
1369 ret = tracing_arm_snapshot_locked(tr);
1370 mutex_unlock(&trace_types_lock);
1371
1372 return ret;
1373}
1374
1375void tracing_disarm_snapshot(struct trace_array *tr)
1376{
1377 spin_lock(&tr->snapshot_trigger_lock);
1378 if (!WARN_ON(!tr->snapshot))
1379 tr->snapshot--;
1380 spin_unlock(&tr->snapshot_trigger_lock);
1381}
1382
1383/**
1384 * tracing_alloc_snapshot - allocate snapshot buffer.
1385 *
1386 * This only allocates the snapshot buffer if it isn't already
1387 * allocated - it doesn't also take a snapshot.
1388 *
1389 * This is meant to be used in cases where the snapshot buffer needs
1390 * to be set up for events that can't sleep but need to be able to
1391 * trigger a snapshot.
1392 */
1393int tracing_alloc_snapshot(void)
1394{
1395 struct trace_array *tr = &global_trace;
1396 int ret;
1397
1398 ret = tracing_alloc_snapshot_instance(tr);
1399 WARN_ON(ret < 0);
1400
1401 return ret;
1402}
1403EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1404
1405/**
1406 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1407 *
1408 * This is similar to tracing_snapshot(), but it will allocate the
1409 * snapshot buffer if it isn't already allocated. Use this only
1410 * where it is safe to sleep, as the allocation may sleep.
1411 *
1412 * This causes a swap between the snapshot buffer and the current live
1413 * tracing buffer. You can use this to take snapshots of the live
1414 * trace when some condition is triggered, but continue to trace.
1415 */
1416void tracing_snapshot_alloc(void)
1417{
1418 int ret;
1419
1420 ret = tracing_alloc_snapshot();
1421 if (ret < 0)
1422 return;
1423
1424 tracing_snapshot();
1425}
1426EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1427
1428/**
1429 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1430 * @tr: The tracing instance
1431 * @cond_data: User data to associate with the snapshot
1432 * @update: Implementation of the cond_snapshot update function
1433 *
1434 * Check whether the conditional snapshot for the given instance has
1435 * already been enabled, or if the current tracer is already using a
1436 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1437 * save the cond_data and update function inside.
1438 *
1439 * Returns 0 if successful, error otherwise.
1440 */
1441int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1442 cond_update_fn_t update)
1443{
1444 struct cond_snapshot *cond_snapshot __free(kfree) =
1445 kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1446 int ret;
1447
1448 if (!cond_snapshot)
1449 return -ENOMEM;
1450
1451 cond_snapshot->cond_data = cond_data;
1452 cond_snapshot->update = update;
1453
1454 guard(mutex)(&trace_types_lock);
1455
1456 if (tr->current_trace->use_max_tr)
1457 return -EBUSY;
1458
1459 /*
1460 * The cond_snapshot can only change to NULL without the
1461 * trace_types_lock. We don't care if we race with it going
1462 * to NULL, but we want to make sure that it's not set to
1463 * something other than NULL when we get here, which we can
1464 * do safely with only holding the trace_types_lock and not
1465 * having to take the max_lock.
1466 */
1467 if (tr->cond_snapshot)
1468 return -EBUSY;
1469
1470 ret = tracing_arm_snapshot_locked(tr);
1471 if (ret)
1472 return ret;
1473
1474 local_irq_disable();
1475 arch_spin_lock(&tr->max_lock);
1476 tr->cond_snapshot = no_free_ptr(cond_snapshot);
1477 arch_spin_unlock(&tr->max_lock);
1478 local_irq_enable();
1479
1480 return 0;
1481}
1482EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1483
1484/**
1485 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1486 * @tr: The tracing instance
1487 *
1488 * Check whether the conditional snapshot for the given instance is
1489 * enabled; if so, free the cond_snapshot associated with it,
1490 * otherwise return -EINVAL.
1491 *
1492 * Returns 0 if successful, error otherwise.
1493 */
1494int tracing_snapshot_cond_disable(struct trace_array *tr)
1495{
1496 int ret = 0;
1497
1498 local_irq_disable();
1499 arch_spin_lock(&tr->max_lock);
1500
1501 if (!tr->cond_snapshot)
1502 ret = -EINVAL;
1503 else {
1504 kfree(tr->cond_snapshot);
1505 tr->cond_snapshot = NULL;
1506 }
1507
1508 arch_spin_unlock(&tr->max_lock);
1509 local_irq_enable();
1510
1511 tracing_disarm_snapshot(tr);
1512
1513 return ret;
1514}
1515EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1516#else
1517void tracing_snapshot(void)
1518{
1519 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1520}
1521EXPORT_SYMBOL_GPL(tracing_snapshot);
1522void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1523{
1524 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1525}
1526EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1527int tracing_alloc_snapshot(void)
1528{
1529 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1530 return -ENODEV;
1531}
1532EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1533void tracing_snapshot_alloc(void)
1534{
1535 /* Give warning */
1536 tracing_snapshot();
1537}
1538EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1539void *tracing_cond_snapshot_data(struct trace_array *tr)
1540{
1541 return NULL;
1542}
1543EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1544int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1545{
1546 return -ENODEV;
1547}
1548EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1549int tracing_snapshot_cond_disable(struct trace_array *tr)
1550{
1551 return false;
1552}
1553EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1554#define free_snapshot(tr) do { } while (0)
1555#define tracing_arm_snapshot_locked(tr) ({ -EBUSY; })
1556#endif /* CONFIG_TRACER_SNAPSHOT */
1557
1558void tracer_tracing_off(struct trace_array *tr)
1559{
1560 if (tr->array_buffer.buffer)
1561 ring_buffer_record_off(tr->array_buffer.buffer);
1562 /*
1563 * This flag is looked at when buffers haven't been allocated
1564 * yet, or by some tracers (like irqsoff), that just want to
1565 * know if the ring buffer has been disabled, but it can handle
1566 * races of where it gets disabled but we still do a record.
1567 * As the check is in the fast path of the tracers, it is more
1568 * important to be fast than accurate.
1569 */
1570 tr->buffer_disabled = 1;
1571 /* Make the flag seen by readers */
1572 smp_wmb();
1573}
1574
1575/**
1576 * tracing_off - turn off tracing buffers
1577 *
1578 * This function stops the tracing buffers from recording data.
1579 * It does not disable any overhead the tracers themselves may
1580 * be causing. This function simply causes all recording to
1581 * the ring buffers to fail.
1582 */
1583void tracing_off(void)
1584{
1585 tracer_tracing_off(&global_trace);
1586}
1587EXPORT_SYMBOL_GPL(tracing_off);
1588
1589void disable_trace_on_warning(void)
1590{
1591 if (__disable_trace_on_warning) {
1592 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1593 "Disabling tracing due to warning\n");
1594 tracing_off();
1595 }
1596}
1597
1598/**
1599 * tracer_tracing_is_on - show real state of ring buffer enabled
1600 * @tr : the trace array to know if ring buffer is enabled
1601 *
1602 * Shows real state of the ring buffer if it is enabled or not.
1603 */
1604bool tracer_tracing_is_on(struct trace_array *tr)
1605{
1606 if (tr->array_buffer.buffer)
1607 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1608 return !tr->buffer_disabled;
1609}
1610
1611/**
1612 * tracing_is_on - show state of ring buffers enabled
1613 */
1614int tracing_is_on(void)
1615{
1616 return tracer_tracing_is_on(&global_trace);
1617}
1618EXPORT_SYMBOL_GPL(tracing_is_on);
1619
1620static int __init set_buf_size(char *str)
1621{
1622 unsigned long buf_size;
1623
1624 if (!str)
1625 return 0;
1626 buf_size = memparse(str, &str);
1627 /*
1628 * nr_entries can not be zero and the startup
1629 * tests require some buffer space. Therefore
1630 * ensure we have at least 4096 bytes of buffer.
1631 */
1632 trace_buf_size = max(4096UL, buf_size);
1633 return 1;
1634}
1635__setup("trace_buf_size=", set_buf_size);
1636
1637static int __init set_tracing_thresh(char *str)
1638{
1639 unsigned long threshold;
1640 int ret;
1641
1642 if (!str)
1643 return 0;
1644 ret = kstrtoul(str, 0, &threshold);
1645 if (ret < 0)
1646 return 0;
1647 tracing_thresh = threshold * 1000;
1648 return 1;
1649}
1650__setup("tracing_thresh=", set_tracing_thresh);
1651
1652unsigned long nsecs_to_usecs(unsigned long nsecs)
1653{
1654 return nsecs / 1000;
1655}
1656
1657/*
1658 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1659 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1660 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1661 * of strings in the order that the evals (enum) were defined.
1662 */
1663#undef C
1664#define C(a, b) b
1665
1666/* These must match the bit positions in trace_iterator_flags */
1667static const char *trace_options[] = {
1668 TRACE_FLAGS
1669 NULL
1670};
1671
1672static struct {
1673 u64 (*func)(void);
1674 const char *name;
1675 int in_ns; /* is this clock in nanoseconds? */
1676} trace_clocks[] = {
1677 { trace_clock_local, "local", 1 },
1678 { trace_clock_global, "global", 1 },
1679 { trace_clock_counter, "counter", 0 },
1680 { trace_clock_jiffies, "uptime", 0 },
1681 { trace_clock, "perf", 1 },
1682 { ktime_get_mono_fast_ns, "mono", 1 },
1683 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1684 { ktime_get_boot_fast_ns, "boot", 1 },
1685 { ktime_get_tai_fast_ns, "tai", 1 },
1686 ARCH_TRACE_CLOCKS
1687};
1688
1689bool trace_clock_in_ns(struct trace_array *tr)
1690{
1691 if (trace_clocks[tr->clock_id].in_ns)
1692 return true;
1693
1694 return false;
1695}
1696
1697/*
1698 * trace_parser_get_init - gets the buffer for trace parser
1699 */
1700int trace_parser_get_init(struct trace_parser *parser, int size)
1701{
1702 memset(parser, 0, sizeof(*parser));
1703
1704 parser->buffer = kmalloc(size, GFP_KERNEL);
1705 if (!parser->buffer)
1706 return 1;
1707
1708 parser->size = size;
1709 return 0;
1710}
1711
1712/*
1713 * trace_parser_put - frees the buffer for trace parser
1714 */
1715void trace_parser_put(struct trace_parser *parser)
1716{
1717 kfree(parser->buffer);
1718 parser->buffer = NULL;
1719}
1720
1721/*
1722 * trace_get_user - reads the user input string separated by space
1723 * (matched by isspace(ch))
1724 *
1725 * For each string found the 'struct trace_parser' is updated,
1726 * and the function returns.
1727 *
1728 * Returns number of bytes read.
1729 *
1730 * See kernel/trace/trace.h for 'struct trace_parser' details.
1731 */
1732int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1733 size_t cnt, loff_t *ppos)
1734{
1735 char ch;
1736 size_t read = 0;
1737 ssize_t ret;
1738
1739 if (!*ppos)
1740 trace_parser_clear(parser);
1741
1742 ret = get_user(ch, ubuf++);
1743 if (ret)
1744 goto out;
1745
1746 read++;
1747 cnt--;
1748
1749 /*
1750 * The parser is not finished with the last write,
1751 * continue reading the user input without skipping spaces.
1752 */
1753 if (!parser->cont) {
1754 /* skip white space */
1755 while (cnt && isspace(ch)) {
1756 ret = get_user(ch, ubuf++);
1757 if (ret)
1758 goto out;
1759 read++;
1760 cnt--;
1761 }
1762
1763 parser->idx = 0;
1764
1765 /* only spaces were written */
1766 if (isspace(ch) || !ch) {
1767 *ppos += read;
1768 ret = read;
1769 goto out;
1770 }
1771 }
1772
1773 /* read the non-space input */
1774 while (cnt && !isspace(ch) && ch) {
1775 if (parser->idx < parser->size - 1)
1776 parser->buffer[parser->idx++] = ch;
1777 else {
1778 ret = -EINVAL;
1779 goto out;
1780 }
1781 ret = get_user(ch, ubuf++);
1782 if (ret)
1783 goto out;
1784 read++;
1785 cnt--;
1786 }
1787
1788 /* We either got finished input or we have to wait for another call. */
1789 if (isspace(ch) || !ch) {
1790 parser->buffer[parser->idx] = 0;
1791 parser->cont = false;
1792 } else if (parser->idx < parser->size - 1) {
1793 parser->cont = true;
1794 parser->buffer[parser->idx++] = ch;
1795 /* Make sure the parsed string always terminates with '\0'. */
1796 parser->buffer[parser->idx] = 0;
1797 } else {
1798 ret = -EINVAL;
1799 goto out;
1800 }
1801
1802 *ppos += read;
1803 ret = read;
1804
1805out:
1806 return ret;
1807}
1808
1809/* TODO add a seq_buf_to_buffer() */
1810static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1811{
1812 int len;
1813
1814 if (trace_seq_used(s) <= s->readpos)
1815 return -EBUSY;
1816
1817 len = trace_seq_used(s) - s->readpos;
1818 if (cnt > len)
1819 cnt = len;
1820 memcpy(buf, s->buffer + s->readpos, cnt);
1821
1822 s->readpos += cnt;
1823 return cnt;
1824}
1825
1826unsigned long __read_mostly tracing_thresh;
1827
1828#ifdef CONFIG_TRACER_MAX_TRACE
1829static const struct file_operations tracing_max_lat_fops;
1830
1831#ifdef LATENCY_FS_NOTIFY
1832
1833static struct workqueue_struct *fsnotify_wq;
1834
1835static void latency_fsnotify_workfn(struct work_struct *work)
1836{
1837 struct trace_array *tr = container_of(work, struct trace_array,
1838 fsnotify_work);
1839 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1840}
1841
1842static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1843{
1844 struct trace_array *tr = container_of(iwork, struct trace_array,
1845 fsnotify_irqwork);
1846 queue_work(fsnotify_wq, &tr->fsnotify_work);
1847}
1848
1849static void trace_create_maxlat_file(struct trace_array *tr,
1850 struct dentry *d_tracer)
1851{
1852 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1853 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1854 tr->d_max_latency = trace_create_file("tracing_max_latency",
1855 TRACE_MODE_WRITE,
1856 d_tracer, tr,
1857 &tracing_max_lat_fops);
1858}
1859
1860__init static int latency_fsnotify_init(void)
1861{
1862 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1863 WQ_UNBOUND | WQ_HIGHPRI, 0);
1864 if (!fsnotify_wq) {
1865 pr_err("Unable to allocate tr_max_lat_wq\n");
1866 return -ENOMEM;
1867 }
1868 return 0;
1869}
1870
1871late_initcall_sync(latency_fsnotify_init);
1872
1873void latency_fsnotify(struct trace_array *tr)
1874{
1875 if (!fsnotify_wq)
1876 return;
1877 /*
1878 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1879 * possible that we are called from __schedule() or do_idle(), which
1880 * could cause a deadlock.
1881 */
1882 irq_work_queue(&tr->fsnotify_irqwork);
1883}
1884
1885#else /* !LATENCY_FS_NOTIFY */
1886
1887#define trace_create_maxlat_file(tr, d_tracer) \
1888 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1889 d_tracer, tr, &tracing_max_lat_fops)
1890
1891#endif
1892
1893/*
1894 * Copy the new maximum trace into the separate maximum-trace
1895 * structure. (this way the maximum trace is permanently saved,
1896 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1897 */
1898static void
1899__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1900{
1901 struct array_buffer *trace_buf = &tr->array_buffer;
1902 struct array_buffer *max_buf = &tr->max_buffer;
1903 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1904 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1905
1906 max_buf->cpu = cpu;
1907 max_buf->time_start = data->preempt_timestamp;
1908
1909 max_data->saved_latency = tr->max_latency;
1910 max_data->critical_start = data->critical_start;
1911 max_data->critical_end = data->critical_end;
1912
1913 strscpy(max_data->comm, tsk->comm);
1914 max_data->pid = tsk->pid;
1915 /*
1916 * If tsk == current, then use current_uid(), as that does not use
1917 * RCU. The irq tracer can be called out of RCU scope.
1918 */
1919 if (tsk == current)
1920 max_data->uid = current_uid();
1921 else
1922 max_data->uid = task_uid(tsk);
1923
1924 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1925 max_data->policy = tsk->policy;
1926 max_data->rt_priority = tsk->rt_priority;
1927
1928 /* record this tasks comm */
1929 tracing_record_cmdline(tsk);
1930 latency_fsnotify(tr);
1931}
1932
1933/**
1934 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1935 * @tr: tracer
1936 * @tsk: the task with the latency
1937 * @cpu: The cpu that initiated the trace.
1938 * @cond_data: User data associated with a conditional snapshot
1939 *
1940 * Flip the buffers between the @tr and the max_tr and record information
1941 * about which task was the cause of this latency.
1942 */
1943void
1944update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1945 void *cond_data)
1946{
1947 if (tr->stop_count)
1948 return;
1949
1950 WARN_ON_ONCE(!irqs_disabled());
1951
1952 if (!tr->allocated_snapshot) {
1953 /* Only the nop tracer should hit this when disabling */
1954 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1955 return;
1956 }
1957
1958 arch_spin_lock(&tr->max_lock);
1959
1960 /* Inherit the recordable setting from array_buffer */
1961 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1962 ring_buffer_record_on(tr->max_buffer.buffer);
1963 else
1964 ring_buffer_record_off(tr->max_buffer.buffer);
1965
1966#ifdef CONFIG_TRACER_SNAPSHOT
1967 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1968 arch_spin_unlock(&tr->max_lock);
1969 return;
1970 }
1971#endif
1972 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1973
1974 __update_max_tr(tr, tsk, cpu);
1975
1976 arch_spin_unlock(&tr->max_lock);
1977
1978 /* Any waiters on the old snapshot buffer need to wake up */
1979 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1980}
1981
1982/**
1983 * update_max_tr_single - only copy one trace over, and reset the rest
1984 * @tr: tracer
1985 * @tsk: task with the latency
1986 * @cpu: the cpu of the buffer to copy.
1987 *
1988 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1989 */
1990void
1991update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1992{
1993 int ret;
1994
1995 if (tr->stop_count)
1996 return;
1997
1998 WARN_ON_ONCE(!irqs_disabled());
1999 if (!tr->allocated_snapshot) {
2000 /* Only the nop tracer should hit this when disabling */
2001 WARN_ON_ONCE(tr->current_trace != &nop_trace);
2002 return;
2003 }
2004
2005 arch_spin_lock(&tr->max_lock);
2006
2007 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
2008
2009 if (ret == -EBUSY) {
2010 /*
2011 * We failed to swap the buffer due to a commit taking
2012 * place on this CPU. We fail to record, but we reset
2013 * the max trace buffer (no one writes directly to it)
2014 * and flag that it failed.
2015 * Another reason is resize is in progress.
2016 */
2017 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
2018 "Failed to swap buffers due to commit or resize in progress\n");
2019 }
2020
2021 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
2022
2023 __update_max_tr(tr, tsk, cpu);
2024 arch_spin_unlock(&tr->max_lock);
2025}
2026
2027#endif /* CONFIG_TRACER_MAX_TRACE */
2028
2029struct pipe_wait {
2030 struct trace_iterator *iter;
2031 int wait_index;
2032};
2033
2034static bool wait_pipe_cond(void *data)
2035{
2036 struct pipe_wait *pwait = data;
2037 struct trace_iterator *iter = pwait->iter;
2038
2039 if (atomic_read_acquire(&iter->wait_index) != pwait->wait_index)
2040 return true;
2041
2042 return iter->closed;
2043}
2044
2045static int wait_on_pipe(struct trace_iterator *iter, int full)
2046{
2047 struct pipe_wait pwait;
2048 int ret;
2049
2050 /* Iterators are static, they should be filled or empty */
2051 if (trace_buffer_iter(iter, iter->cpu_file))
2052 return 0;
2053
2054 pwait.wait_index = atomic_read_acquire(&iter->wait_index);
2055 pwait.iter = iter;
2056
2057 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full,
2058 wait_pipe_cond, &pwait);
2059
2060#ifdef CONFIG_TRACER_MAX_TRACE
2061 /*
2062 * Make sure this is still the snapshot buffer, as if a snapshot were
2063 * to happen, this would now be the main buffer.
2064 */
2065 if (iter->snapshot)
2066 iter->array_buffer = &iter->tr->max_buffer;
2067#endif
2068 return ret;
2069}
2070
2071#ifdef CONFIG_FTRACE_STARTUP_TEST
2072static bool selftests_can_run;
2073
2074struct trace_selftests {
2075 struct list_head list;
2076 struct tracer *type;
2077};
2078
2079static LIST_HEAD(postponed_selftests);
2080
2081static int save_selftest(struct tracer *type)
2082{
2083 struct trace_selftests *selftest;
2084
2085 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
2086 if (!selftest)
2087 return -ENOMEM;
2088
2089 selftest->type = type;
2090 list_add(&selftest->list, &postponed_selftests);
2091 return 0;
2092}
2093
2094static int run_tracer_selftest(struct tracer *type)
2095{
2096 struct trace_array *tr = &global_trace;
2097 struct tracer *saved_tracer = tr->current_trace;
2098 int ret;
2099
2100 if (!type->selftest || tracing_selftest_disabled)
2101 return 0;
2102
2103 /*
2104 * If a tracer registers early in boot up (before scheduling is
2105 * initialized and such), then do not run its selftests yet.
2106 * Instead, run it a little later in the boot process.
2107 */
2108 if (!selftests_can_run)
2109 return save_selftest(type);
2110
2111 if (!tracing_is_on()) {
2112 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2113 type->name);
2114 return 0;
2115 }
2116
2117 /*
2118 * Run a selftest on this tracer.
2119 * Here we reset the trace buffer, and set the current
2120 * tracer to be this tracer. The tracer can then run some
2121 * internal tracing to verify that everything is in order.
2122 * If we fail, we do not register this tracer.
2123 */
2124 tracing_reset_online_cpus(&tr->array_buffer);
2125
2126 tr->current_trace = type;
2127
2128#ifdef CONFIG_TRACER_MAX_TRACE
2129 if (type->use_max_tr) {
2130 /* If we expanded the buffers, make sure the max is expanded too */
2131 if (tr->ring_buffer_expanded)
2132 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2133 RING_BUFFER_ALL_CPUS);
2134 tr->allocated_snapshot = true;
2135 }
2136#endif
2137
2138 /* the test is responsible for initializing and enabling */
2139 pr_info("Testing tracer %s: ", type->name);
2140 ret = type->selftest(type, tr);
2141 /* the test is responsible for resetting too */
2142 tr->current_trace = saved_tracer;
2143 if (ret) {
2144 printk(KERN_CONT "FAILED!\n");
2145 /* Add the warning after printing 'FAILED' */
2146 WARN_ON(1);
2147 return -1;
2148 }
2149 /* Only reset on passing, to avoid touching corrupted buffers */
2150 tracing_reset_online_cpus(&tr->array_buffer);
2151
2152#ifdef CONFIG_TRACER_MAX_TRACE
2153 if (type->use_max_tr) {
2154 tr->allocated_snapshot = false;
2155
2156 /* Shrink the max buffer again */
2157 if (tr->ring_buffer_expanded)
2158 ring_buffer_resize(tr->max_buffer.buffer, 1,
2159 RING_BUFFER_ALL_CPUS);
2160 }
2161#endif
2162
2163 printk(KERN_CONT "PASSED\n");
2164 return 0;
2165}
2166
2167static int do_run_tracer_selftest(struct tracer *type)
2168{
2169 int ret;
2170
2171 /*
2172 * Tests can take a long time, especially if they are run one after the
2173 * other, as does happen during bootup when all the tracers are
2174 * registered. This could cause the soft lockup watchdog to trigger.
2175 */
2176 cond_resched();
2177
2178 tracing_selftest_running = true;
2179 ret = run_tracer_selftest(type);
2180 tracing_selftest_running = false;
2181
2182 return ret;
2183}
2184
2185static __init int init_trace_selftests(void)
2186{
2187 struct trace_selftests *p, *n;
2188 struct tracer *t, **last;
2189 int ret;
2190
2191 selftests_can_run = true;
2192
2193 guard(mutex)(&trace_types_lock);
2194
2195 if (list_empty(&postponed_selftests))
2196 return 0;
2197
2198 pr_info("Running postponed tracer tests:\n");
2199
2200 tracing_selftest_running = true;
2201 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2202 /* This loop can take minutes when sanitizers are enabled, so
2203 * lets make sure we allow RCU processing.
2204 */
2205 cond_resched();
2206 ret = run_tracer_selftest(p->type);
2207 /* If the test fails, then warn and remove from available_tracers */
2208 if (ret < 0) {
2209 WARN(1, "tracer: %s failed selftest, disabling\n",
2210 p->type->name);
2211 last = &trace_types;
2212 for (t = trace_types; t; t = t->next) {
2213 if (t == p->type) {
2214 *last = t->next;
2215 break;
2216 }
2217 last = &t->next;
2218 }
2219 }
2220 list_del(&p->list);
2221 kfree(p);
2222 }
2223 tracing_selftest_running = false;
2224
2225 return 0;
2226}
2227core_initcall(init_trace_selftests);
2228#else
2229static inline int do_run_tracer_selftest(struct tracer *type)
2230{
2231 return 0;
2232}
2233#endif /* CONFIG_FTRACE_STARTUP_TEST */
2234
2235static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2236
2237static void __init apply_trace_boot_options(void);
2238
2239/**
2240 * register_tracer - register a tracer with the ftrace system.
2241 * @type: the plugin for the tracer
2242 *
2243 * Register a new plugin tracer.
2244 */
2245int __init register_tracer(struct tracer *type)
2246{
2247 struct tracer *t;
2248 int ret = 0;
2249
2250 if (!type->name) {
2251 pr_info("Tracer must have a name\n");
2252 return -1;
2253 }
2254
2255 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2256 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2257 return -1;
2258 }
2259
2260 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2261 pr_warn("Can not register tracer %s due to lockdown\n",
2262 type->name);
2263 return -EPERM;
2264 }
2265
2266 mutex_lock(&trace_types_lock);
2267
2268 for (t = trace_types; t; t = t->next) {
2269 if (strcmp(type->name, t->name) == 0) {
2270 /* already found */
2271 pr_info("Tracer %s already registered\n",
2272 type->name);
2273 ret = -1;
2274 goto out;
2275 }
2276 }
2277
2278 if (!type->set_flag)
2279 type->set_flag = &dummy_set_flag;
2280 if (!type->flags) {
2281 /*allocate a dummy tracer_flags*/
2282 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2283 if (!type->flags) {
2284 ret = -ENOMEM;
2285 goto out;
2286 }
2287 type->flags->val = 0;
2288 type->flags->opts = dummy_tracer_opt;
2289 } else
2290 if (!type->flags->opts)
2291 type->flags->opts = dummy_tracer_opt;
2292
2293 /* store the tracer for __set_tracer_option */
2294 type->flags->trace = type;
2295
2296 ret = do_run_tracer_selftest(type);
2297 if (ret < 0)
2298 goto out;
2299
2300 type->next = trace_types;
2301 trace_types = type;
2302 add_tracer_options(&global_trace, type);
2303
2304 out:
2305 mutex_unlock(&trace_types_lock);
2306
2307 if (ret || !default_bootup_tracer)
2308 goto out_unlock;
2309
2310 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2311 goto out_unlock;
2312
2313 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2314 /* Do we want this tracer to start on bootup? */
2315 tracing_set_tracer(&global_trace, type->name);
2316 default_bootup_tracer = NULL;
2317
2318 apply_trace_boot_options();
2319
2320 /* disable other selftests, since this will break it. */
2321 disable_tracing_selftest("running a tracer");
2322
2323 out_unlock:
2324 return ret;
2325}
2326
2327static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2328{
2329 struct trace_buffer *buffer = buf->buffer;
2330
2331 if (!buffer)
2332 return;
2333
2334 ring_buffer_record_disable(buffer);
2335
2336 /* Make sure all commits have finished */
2337 synchronize_rcu();
2338 ring_buffer_reset_cpu(buffer, cpu);
2339
2340 ring_buffer_record_enable(buffer);
2341}
2342
2343void tracing_reset_online_cpus(struct array_buffer *buf)
2344{
2345 struct trace_buffer *buffer = buf->buffer;
2346
2347 if (!buffer)
2348 return;
2349
2350 ring_buffer_record_disable(buffer);
2351
2352 /* Make sure all commits have finished */
2353 synchronize_rcu();
2354
2355 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2356
2357 ring_buffer_reset_online_cpus(buffer);
2358
2359 ring_buffer_record_enable(buffer);
2360}
2361
2362static void tracing_reset_all_cpus(struct array_buffer *buf)
2363{
2364 struct trace_buffer *buffer = buf->buffer;
2365
2366 if (!buffer)
2367 return;
2368
2369 ring_buffer_record_disable(buffer);
2370
2371 /* Make sure all commits have finished */
2372 synchronize_rcu();
2373
2374 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2375
2376 ring_buffer_reset(buffer);
2377
2378 ring_buffer_record_enable(buffer);
2379}
2380
2381/* Must have trace_types_lock held */
2382void tracing_reset_all_online_cpus_unlocked(void)
2383{
2384 struct trace_array *tr;
2385
2386 lockdep_assert_held(&trace_types_lock);
2387
2388 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2389 if (!tr->clear_trace)
2390 continue;
2391 tr->clear_trace = false;
2392 tracing_reset_online_cpus(&tr->array_buffer);
2393#ifdef CONFIG_TRACER_MAX_TRACE
2394 tracing_reset_online_cpus(&tr->max_buffer);
2395#endif
2396 }
2397}
2398
2399void tracing_reset_all_online_cpus(void)
2400{
2401 mutex_lock(&trace_types_lock);
2402 tracing_reset_all_online_cpus_unlocked();
2403 mutex_unlock(&trace_types_lock);
2404}
2405
2406int is_tracing_stopped(void)
2407{
2408 return global_trace.stop_count;
2409}
2410
2411static void tracing_start_tr(struct trace_array *tr)
2412{
2413 struct trace_buffer *buffer;
2414 unsigned long flags;
2415
2416 if (tracing_disabled)
2417 return;
2418
2419 raw_spin_lock_irqsave(&tr->start_lock, flags);
2420 if (--tr->stop_count) {
2421 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2422 /* Someone screwed up their debugging */
2423 tr->stop_count = 0;
2424 }
2425 goto out;
2426 }
2427
2428 /* Prevent the buffers from switching */
2429 arch_spin_lock(&tr->max_lock);
2430
2431 buffer = tr->array_buffer.buffer;
2432 if (buffer)
2433 ring_buffer_record_enable(buffer);
2434
2435#ifdef CONFIG_TRACER_MAX_TRACE
2436 buffer = tr->max_buffer.buffer;
2437 if (buffer)
2438 ring_buffer_record_enable(buffer);
2439#endif
2440
2441 arch_spin_unlock(&tr->max_lock);
2442
2443 out:
2444 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2445}
2446
2447/**
2448 * tracing_start - quick start of the tracer
2449 *
2450 * If tracing is enabled but was stopped by tracing_stop,
2451 * this will start the tracer back up.
2452 */
2453void tracing_start(void)
2454
2455{
2456 return tracing_start_tr(&global_trace);
2457}
2458
2459static void tracing_stop_tr(struct trace_array *tr)
2460{
2461 struct trace_buffer *buffer;
2462 unsigned long flags;
2463
2464 raw_spin_lock_irqsave(&tr->start_lock, flags);
2465 if (tr->stop_count++)
2466 goto out;
2467
2468 /* Prevent the buffers from switching */
2469 arch_spin_lock(&tr->max_lock);
2470
2471 buffer = tr->array_buffer.buffer;
2472 if (buffer)
2473 ring_buffer_record_disable(buffer);
2474
2475#ifdef CONFIG_TRACER_MAX_TRACE
2476 buffer = tr->max_buffer.buffer;
2477 if (buffer)
2478 ring_buffer_record_disable(buffer);
2479#endif
2480
2481 arch_spin_unlock(&tr->max_lock);
2482
2483 out:
2484 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2485}
2486
2487/**
2488 * tracing_stop - quick stop of the tracer
2489 *
2490 * Light weight way to stop tracing. Use in conjunction with
2491 * tracing_start.
2492 */
2493void tracing_stop(void)
2494{
2495 return tracing_stop_tr(&global_trace);
2496}
2497
2498/*
2499 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2500 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2501 * simplifies those functions and keeps them in sync.
2502 */
2503enum print_line_t trace_handle_return(struct trace_seq *s)
2504{
2505 return trace_seq_has_overflowed(s) ?
2506 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2507}
2508EXPORT_SYMBOL_GPL(trace_handle_return);
2509
2510static unsigned short migration_disable_value(void)
2511{
2512#if defined(CONFIG_SMP)
2513 return current->migration_disabled;
2514#else
2515 return 0;
2516#endif
2517}
2518
2519unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2520{
2521 unsigned int trace_flags = irqs_status;
2522 unsigned int pc;
2523
2524 pc = preempt_count();
2525
2526 if (pc & NMI_MASK)
2527 trace_flags |= TRACE_FLAG_NMI;
2528 if (pc & HARDIRQ_MASK)
2529 trace_flags |= TRACE_FLAG_HARDIRQ;
2530 if (in_serving_softirq())
2531 trace_flags |= TRACE_FLAG_SOFTIRQ;
2532 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2533 trace_flags |= TRACE_FLAG_BH_OFF;
2534
2535 if (tif_need_resched())
2536 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2537 if (test_preempt_need_resched())
2538 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2539 if (IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY) && tif_test_bit(TIF_NEED_RESCHED_LAZY))
2540 trace_flags |= TRACE_FLAG_NEED_RESCHED_LAZY;
2541 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2542 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2543}
2544
2545struct ring_buffer_event *
2546trace_buffer_lock_reserve(struct trace_buffer *buffer,
2547 int type,
2548 unsigned long len,
2549 unsigned int trace_ctx)
2550{
2551 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2552}
2553
2554DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2555DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2556static int trace_buffered_event_ref;
2557
2558/**
2559 * trace_buffered_event_enable - enable buffering events
2560 *
2561 * When events are being filtered, it is quicker to use a temporary
2562 * buffer to write the event data into if there's a likely chance
2563 * that it will not be committed. The discard of the ring buffer
2564 * is not as fast as committing, and is much slower than copying
2565 * a commit.
2566 *
2567 * When an event is to be filtered, allocate per cpu buffers to
2568 * write the event data into, and if the event is filtered and discarded
2569 * it is simply dropped, otherwise, the entire data is to be committed
2570 * in one shot.
2571 */
2572void trace_buffered_event_enable(void)
2573{
2574 struct ring_buffer_event *event;
2575 struct page *page;
2576 int cpu;
2577
2578 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2579
2580 if (trace_buffered_event_ref++)
2581 return;
2582
2583 for_each_tracing_cpu(cpu) {
2584 page = alloc_pages_node(cpu_to_node(cpu),
2585 GFP_KERNEL | __GFP_NORETRY, 0);
2586 /* This is just an optimization and can handle failures */
2587 if (!page) {
2588 pr_err("Failed to allocate event buffer\n");
2589 break;
2590 }
2591
2592 event = page_address(page);
2593 memset(event, 0, sizeof(*event));
2594
2595 per_cpu(trace_buffered_event, cpu) = event;
2596
2597 preempt_disable();
2598 if (cpu == smp_processor_id() &&
2599 __this_cpu_read(trace_buffered_event) !=
2600 per_cpu(trace_buffered_event, cpu))
2601 WARN_ON_ONCE(1);
2602 preempt_enable();
2603 }
2604}
2605
2606static void enable_trace_buffered_event(void *data)
2607{
2608 /* Probably not needed, but do it anyway */
2609 smp_rmb();
2610 this_cpu_dec(trace_buffered_event_cnt);
2611}
2612
2613static void disable_trace_buffered_event(void *data)
2614{
2615 this_cpu_inc(trace_buffered_event_cnt);
2616}
2617
2618/**
2619 * trace_buffered_event_disable - disable buffering events
2620 *
2621 * When a filter is removed, it is faster to not use the buffered
2622 * events, and to commit directly into the ring buffer. Free up
2623 * the temp buffers when there are no more users. This requires
2624 * special synchronization with current events.
2625 */
2626void trace_buffered_event_disable(void)
2627{
2628 int cpu;
2629
2630 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2631
2632 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2633 return;
2634
2635 if (--trace_buffered_event_ref)
2636 return;
2637
2638 /* For each CPU, set the buffer as used. */
2639 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2640 NULL, true);
2641
2642 /* Wait for all current users to finish */
2643 synchronize_rcu();
2644
2645 for_each_tracing_cpu(cpu) {
2646 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2647 per_cpu(trace_buffered_event, cpu) = NULL;
2648 }
2649
2650 /*
2651 * Wait for all CPUs that potentially started checking if they can use
2652 * their event buffer only after the previous synchronize_rcu() call and
2653 * they still read a valid pointer from trace_buffered_event. It must be
2654 * ensured they don't see cleared trace_buffered_event_cnt else they
2655 * could wrongly decide to use the pointed-to buffer which is now freed.
2656 */
2657 synchronize_rcu();
2658
2659 /* For each CPU, relinquish the buffer */
2660 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2661 true);
2662}
2663
2664static struct trace_buffer *temp_buffer;
2665
2666struct ring_buffer_event *
2667trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2668 struct trace_event_file *trace_file,
2669 int type, unsigned long len,
2670 unsigned int trace_ctx)
2671{
2672 struct ring_buffer_event *entry;
2673 struct trace_array *tr = trace_file->tr;
2674 int val;
2675
2676 *current_rb = tr->array_buffer.buffer;
2677
2678 if (!tr->no_filter_buffering_ref &&
2679 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2680 preempt_disable_notrace();
2681 /*
2682 * Filtering is on, so try to use the per cpu buffer first.
2683 * This buffer will simulate a ring_buffer_event,
2684 * where the type_len is zero and the array[0] will
2685 * hold the full length.
2686 * (see include/linux/ring-buffer.h for details on
2687 * how the ring_buffer_event is structured).
2688 *
2689 * Using a temp buffer during filtering and copying it
2690 * on a matched filter is quicker than writing directly
2691 * into the ring buffer and then discarding it when
2692 * it doesn't match. That is because the discard
2693 * requires several atomic operations to get right.
2694 * Copying on match and doing nothing on a failed match
2695 * is still quicker than no copy on match, but having
2696 * to discard out of the ring buffer on a failed match.
2697 */
2698 if ((entry = __this_cpu_read(trace_buffered_event))) {
2699 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2700
2701 val = this_cpu_inc_return(trace_buffered_event_cnt);
2702
2703 /*
2704 * Preemption is disabled, but interrupts and NMIs
2705 * can still come in now. If that happens after
2706 * the above increment, then it will have to go
2707 * back to the old method of allocating the event
2708 * on the ring buffer, and if the filter fails, it
2709 * will have to call ring_buffer_discard_commit()
2710 * to remove it.
2711 *
2712 * Need to also check the unlikely case that the
2713 * length is bigger than the temp buffer size.
2714 * If that happens, then the reserve is pretty much
2715 * guaranteed to fail, as the ring buffer currently
2716 * only allows events less than a page. But that may
2717 * change in the future, so let the ring buffer reserve
2718 * handle the failure in that case.
2719 */
2720 if (val == 1 && likely(len <= max_len)) {
2721 trace_event_setup(entry, type, trace_ctx);
2722 entry->array[0] = len;
2723 /* Return with preemption disabled */
2724 return entry;
2725 }
2726 this_cpu_dec(trace_buffered_event_cnt);
2727 }
2728 /* __trace_buffer_lock_reserve() disables preemption */
2729 preempt_enable_notrace();
2730 }
2731
2732 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2733 trace_ctx);
2734 /*
2735 * If tracing is off, but we have triggers enabled
2736 * we still need to look at the event data. Use the temp_buffer
2737 * to store the trace event for the trigger to use. It's recursive
2738 * safe and will not be recorded anywhere.
2739 */
2740 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2741 *current_rb = temp_buffer;
2742 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2743 trace_ctx);
2744 }
2745 return entry;
2746}
2747EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2748
2749static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2750static DEFINE_MUTEX(tracepoint_printk_mutex);
2751
2752static void output_printk(struct trace_event_buffer *fbuffer)
2753{
2754 struct trace_event_call *event_call;
2755 struct trace_event_file *file;
2756 struct trace_event *event;
2757 unsigned long flags;
2758 struct trace_iterator *iter = tracepoint_print_iter;
2759
2760 /* We should never get here if iter is NULL */
2761 if (WARN_ON_ONCE(!iter))
2762 return;
2763
2764 event_call = fbuffer->trace_file->event_call;
2765 if (!event_call || !event_call->event.funcs ||
2766 !event_call->event.funcs->trace)
2767 return;
2768
2769 file = fbuffer->trace_file;
2770 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2771 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2772 !filter_match_preds(file->filter, fbuffer->entry)))
2773 return;
2774
2775 event = &fbuffer->trace_file->event_call->event;
2776
2777 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2778 trace_seq_init(&iter->seq);
2779 iter->ent = fbuffer->entry;
2780 event_call->event.funcs->trace(iter, 0, event);
2781 trace_seq_putc(&iter->seq, 0);
2782 printk("%s", iter->seq.buffer);
2783
2784 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2785}
2786
2787int tracepoint_printk_sysctl(const struct ctl_table *table, int write,
2788 void *buffer, size_t *lenp,
2789 loff_t *ppos)
2790{
2791 int save_tracepoint_printk;
2792 int ret;
2793
2794 guard(mutex)(&tracepoint_printk_mutex);
2795 save_tracepoint_printk = tracepoint_printk;
2796
2797 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2798
2799 /*
2800 * This will force exiting early, as tracepoint_printk
2801 * is always zero when tracepoint_printk_iter is not allocated
2802 */
2803 if (!tracepoint_print_iter)
2804 tracepoint_printk = 0;
2805
2806 if (save_tracepoint_printk == tracepoint_printk)
2807 return ret;
2808
2809 if (tracepoint_printk)
2810 static_key_enable(&tracepoint_printk_key.key);
2811 else
2812 static_key_disable(&tracepoint_printk_key.key);
2813
2814 return ret;
2815}
2816
2817void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2818{
2819 enum event_trigger_type tt = ETT_NONE;
2820 struct trace_event_file *file = fbuffer->trace_file;
2821
2822 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2823 fbuffer->entry, &tt))
2824 goto discard;
2825
2826 if (static_key_false(&tracepoint_printk_key.key))
2827 output_printk(fbuffer);
2828
2829 if (static_branch_unlikely(&trace_event_exports_enabled))
2830 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2831
2832 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2833 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2834
2835discard:
2836 if (tt)
2837 event_triggers_post_call(file, tt);
2838
2839}
2840EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2841
2842/*
2843 * Skip 3:
2844 *
2845 * trace_buffer_unlock_commit_regs()
2846 * trace_event_buffer_commit()
2847 * trace_event_raw_event_xxx()
2848 */
2849# define STACK_SKIP 3
2850
2851void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2852 struct trace_buffer *buffer,
2853 struct ring_buffer_event *event,
2854 unsigned int trace_ctx,
2855 struct pt_regs *regs)
2856{
2857 __buffer_unlock_commit(buffer, event);
2858
2859 /*
2860 * If regs is not set, then skip the necessary functions.
2861 * Note, we can still get here via blktrace, wakeup tracer
2862 * and mmiotrace, but that's ok if they lose a function or
2863 * two. They are not that meaningful.
2864 */
2865 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2866 ftrace_trace_userstack(tr, buffer, trace_ctx);
2867}
2868
2869/*
2870 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2871 */
2872void
2873trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2874 struct ring_buffer_event *event)
2875{
2876 __buffer_unlock_commit(buffer, event);
2877}
2878
2879void
2880trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2881 parent_ip, unsigned int trace_ctx)
2882{
2883 struct trace_buffer *buffer = tr->array_buffer.buffer;
2884 struct ring_buffer_event *event;
2885 struct ftrace_entry *entry;
2886
2887 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2888 trace_ctx);
2889 if (!event)
2890 return;
2891 entry = ring_buffer_event_data(event);
2892 entry->ip = ip;
2893 entry->parent_ip = parent_ip;
2894
2895 if (static_branch_unlikely(&trace_function_exports_enabled))
2896 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2897 __buffer_unlock_commit(buffer, event);
2898}
2899
2900#ifdef CONFIG_STACKTRACE
2901
2902/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2903#define FTRACE_KSTACK_NESTING 4
2904
2905#define FTRACE_KSTACK_ENTRIES (SZ_4K / FTRACE_KSTACK_NESTING)
2906
2907struct ftrace_stack {
2908 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2909};
2910
2911
2912struct ftrace_stacks {
2913 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2914};
2915
2916static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2917static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2918
2919static void __ftrace_trace_stack(struct trace_array *tr,
2920 struct trace_buffer *buffer,
2921 unsigned int trace_ctx,
2922 int skip, struct pt_regs *regs)
2923{
2924 struct ring_buffer_event *event;
2925 unsigned int size, nr_entries;
2926 struct ftrace_stack *fstack;
2927 struct stack_entry *entry;
2928 int stackidx;
2929
2930 /*
2931 * Add one, for this function and the call to save_stack_trace()
2932 * If regs is set, then these functions will not be in the way.
2933 */
2934#ifndef CONFIG_UNWINDER_ORC
2935 if (!regs)
2936 skip++;
2937#endif
2938
2939 preempt_disable_notrace();
2940
2941 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2942
2943 /* This should never happen. If it does, yell once and skip */
2944 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2945 goto out;
2946
2947 /*
2948 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2949 * interrupt will either see the value pre increment or post
2950 * increment. If the interrupt happens pre increment it will have
2951 * restored the counter when it returns. We just need a barrier to
2952 * keep gcc from moving things around.
2953 */
2954 barrier();
2955
2956 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2957 size = ARRAY_SIZE(fstack->calls);
2958
2959 if (regs) {
2960 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2961 size, skip);
2962 } else {
2963 nr_entries = stack_trace_save(fstack->calls, size, skip);
2964 }
2965
2966#ifdef CONFIG_DYNAMIC_FTRACE
2967 /* Mark entry of stack trace as trampoline code */
2968 if (tr->ops && tr->ops->trampoline) {
2969 unsigned long tramp_start = tr->ops->trampoline;
2970 unsigned long tramp_end = tramp_start + tr->ops->trampoline_size;
2971 unsigned long *calls = fstack->calls;
2972
2973 for (int i = 0; i < nr_entries; i++) {
2974 if (calls[i] >= tramp_start && calls[i] < tramp_end)
2975 calls[i] = FTRACE_TRAMPOLINE_MARKER;
2976 }
2977 }
2978#endif
2979
2980 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2981 struct_size(entry, caller, nr_entries),
2982 trace_ctx);
2983 if (!event)
2984 goto out;
2985 entry = ring_buffer_event_data(event);
2986
2987 entry->size = nr_entries;
2988 memcpy(&entry->caller, fstack->calls,
2989 flex_array_size(entry, caller, nr_entries));
2990
2991 __buffer_unlock_commit(buffer, event);
2992
2993 out:
2994 /* Again, don't let gcc optimize things here */
2995 barrier();
2996 __this_cpu_dec(ftrace_stack_reserve);
2997 preempt_enable_notrace();
2998
2999}
3000
3001static inline void ftrace_trace_stack(struct trace_array *tr,
3002 struct trace_buffer *buffer,
3003 unsigned int trace_ctx,
3004 int skip, struct pt_regs *regs)
3005{
3006 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3007 return;
3008
3009 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, regs);
3010}
3011
3012void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3013 int skip)
3014{
3015 struct trace_buffer *buffer = tr->array_buffer.buffer;
3016
3017 if (rcu_is_watching()) {
3018 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, NULL);
3019 return;
3020 }
3021
3022 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3023 return;
3024
3025 /*
3026 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3027 * but if the above rcu_is_watching() failed, then the NMI
3028 * triggered someplace critical, and ct_irq_enter() should
3029 * not be called from NMI.
3030 */
3031 if (unlikely(in_nmi()))
3032 return;
3033
3034 ct_irq_enter_irqson();
3035 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, NULL);
3036 ct_irq_exit_irqson();
3037}
3038
3039/**
3040 * trace_dump_stack - record a stack back trace in the trace buffer
3041 * @skip: Number of functions to skip (helper handlers)
3042 */
3043void trace_dump_stack(int skip)
3044{
3045 if (tracing_disabled || tracing_selftest_running)
3046 return;
3047
3048#ifndef CONFIG_UNWINDER_ORC
3049 /* Skip 1 to skip this function. */
3050 skip++;
3051#endif
3052 __ftrace_trace_stack(printk_trace, printk_trace->array_buffer.buffer,
3053 tracing_gen_ctx(), skip, NULL);
3054}
3055EXPORT_SYMBOL_GPL(trace_dump_stack);
3056
3057#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3058static DEFINE_PER_CPU(int, user_stack_count);
3059
3060static void
3061ftrace_trace_userstack(struct trace_array *tr,
3062 struct trace_buffer *buffer, unsigned int trace_ctx)
3063{
3064 struct ring_buffer_event *event;
3065 struct userstack_entry *entry;
3066
3067 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3068 return;
3069
3070 /*
3071 * NMIs can not handle page faults, even with fix ups.
3072 * The save user stack can (and often does) fault.
3073 */
3074 if (unlikely(in_nmi()))
3075 return;
3076
3077 /*
3078 * prevent recursion, since the user stack tracing may
3079 * trigger other kernel events.
3080 */
3081 preempt_disable();
3082 if (__this_cpu_read(user_stack_count))
3083 goto out;
3084
3085 __this_cpu_inc(user_stack_count);
3086
3087 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3088 sizeof(*entry), trace_ctx);
3089 if (!event)
3090 goto out_drop_count;
3091 entry = ring_buffer_event_data(event);
3092
3093 entry->tgid = current->tgid;
3094 memset(&entry->caller, 0, sizeof(entry->caller));
3095
3096 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3097 __buffer_unlock_commit(buffer, event);
3098
3099 out_drop_count:
3100 __this_cpu_dec(user_stack_count);
3101 out:
3102 preempt_enable();
3103}
3104#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3105static void ftrace_trace_userstack(struct trace_array *tr,
3106 struct trace_buffer *buffer,
3107 unsigned int trace_ctx)
3108{
3109}
3110#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3111
3112#endif /* CONFIG_STACKTRACE */
3113
3114static inline void
3115func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3116 unsigned long long delta)
3117{
3118 entry->bottom_delta_ts = delta & U32_MAX;
3119 entry->top_delta_ts = (delta >> 32);
3120}
3121
3122void trace_last_func_repeats(struct trace_array *tr,
3123 struct trace_func_repeats *last_info,
3124 unsigned int trace_ctx)
3125{
3126 struct trace_buffer *buffer = tr->array_buffer.buffer;
3127 struct func_repeats_entry *entry;
3128 struct ring_buffer_event *event;
3129 u64 delta;
3130
3131 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3132 sizeof(*entry), trace_ctx);
3133 if (!event)
3134 return;
3135
3136 delta = ring_buffer_event_time_stamp(buffer, event) -
3137 last_info->ts_last_call;
3138
3139 entry = ring_buffer_event_data(event);
3140 entry->ip = last_info->ip;
3141 entry->parent_ip = last_info->parent_ip;
3142 entry->count = last_info->count;
3143 func_repeats_set_delta_ts(entry, delta);
3144
3145 __buffer_unlock_commit(buffer, event);
3146}
3147
3148/* created for use with alloc_percpu */
3149struct trace_buffer_struct {
3150 int nesting;
3151 char buffer[4][TRACE_BUF_SIZE];
3152};
3153
3154static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3155
3156/*
3157 * This allows for lockless recording. If we're nested too deeply, then
3158 * this returns NULL.
3159 */
3160static char *get_trace_buf(void)
3161{
3162 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3163
3164 if (!trace_percpu_buffer || buffer->nesting >= 4)
3165 return NULL;
3166
3167 buffer->nesting++;
3168
3169 /* Interrupts must see nesting incremented before we use the buffer */
3170 barrier();
3171 return &buffer->buffer[buffer->nesting - 1][0];
3172}
3173
3174static void put_trace_buf(void)
3175{
3176 /* Don't let the decrement of nesting leak before this */
3177 barrier();
3178 this_cpu_dec(trace_percpu_buffer->nesting);
3179}
3180
3181static int alloc_percpu_trace_buffer(void)
3182{
3183 struct trace_buffer_struct __percpu *buffers;
3184
3185 if (trace_percpu_buffer)
3186 return 0;
3187
3188 buffers = alloc_percpu(struct trace_buffer_struct);
3189 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3190 return -ENOMEM;
3191
3192 trace_percpu_buffer = buffers;
3193 return 0;
3194}
3195
3196static int buffers_allocated;
3197
3198void trace_printk_init_buffers(void)
3199{
3200 if (buffers_allocated)
3201 return;
3202
3203 if (alloc_percpu_trace_buffer())
3204 return;
3205
3206 /* trace_printk() is for debug use only. Don't use it in production. */
3207
3208 pr_warn("\n");
3209 pr_warn("**********************************************************\n");
3210 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3211 pr_warn("** **\n");
3212 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3213 pr_warn("** **\n");
3214 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3215 pr_warn("** unsafe for production use. **\n");
3216 pr_warn("** **\n");
3217 pr_warn("** If you see this message and you are not debugging **\n");
3218 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3219 pr_warn("** **\n");
3220 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3221 pr_warn("**********************************************************\n");
3222
3223 /* Expand the buffers to set size */
3224 tracing_update_buffers(&global_trace);
3225
3226 buffers_allocated = 1;
3227
3228 /*
3229 * trace_printk_init_buffers() can be called by modules.
3230 * If that happens, then we need to start cmdline recording
3231 * directly here. If the global_trace.buffer is already
3232 * allocated here, then this was called by module code.
3233 */
3234 if (global_trace.array_buffer.buffer)
3235 tracing_start_cmdline_record();
3236}
3237EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3238
3239void trace_printk_start_comm(void)
3240{
3241 /* Start tracing comms if trace printk is set */
3242 if (!buffers_allocated)
3243 return;
3244 tracing_start_cmdline_record();
3245}
3246
3247static void trace_printk_start_stop_comm(int enabled)
3248{
3249 if (!buffers_allocated)
3250 return;
3251
3252 if (enabled)
3253 tracing_start_cmdline_record();
3254 else
3255 tracing_stop_cmdline_record();
3256}
3257
3258/**
3259 * trace_vbprintk - write binary msg to tracing buffer
3260 * @ip: The address of the caller
3261 * @fmt: The string format to write to the buffer
3262 * @args: Arguments for @fmt
3263 */
3264int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3265{
3266 struct ring_buffer_event *event;
3267 struct trace_buffer *buffer;
3268 struct trace_array *tr = READ_ONCE(printk_trace);
3269 struct bprint_entry *entry;
3270 unsigned int trace_ctx;
3271 char *tbuffer;
3272 int len = 0, size;
3273
3274 if (!printk_binsafe(tr))
3275 return trace_vprintk(ip, fmt, args);
3276
3277 if (unlikely(tracing_selftest_running || tracing_disabled))
3278 return 0;
3279
3280 /* Don't pollute graph traces with trace_vprintk internals */
3281 pause_graph_tracing();
3282
3283 trace_ctx = tracing_gen_ctx();
3284 preempt_disable_notrace();
3285
3286 tbuffer = get_trace_buf();
3287 if (!tbuffer) {
3288 len = 0;
3289 goto out_nobuffer;
3290 }
3291
3292 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3293
3294 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3295 goto out_put;
3296
3297 size = sizeof(*entry) + sizeof(u32) * len;
3298 buffer = tr->array_buffer.buffer;
3299 ring_buffer_nest_start(buffer);
3300 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3301 trace_ctx);
3302 if (!event)
3303 goto out;
3304 entry = ring_buffer_event_data(event);
3305 entry->ip = ip;
3306 entry->fmt = fmt;
3307
3308 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3309 __buffer_unlock_commit(buffer, event);
3310 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3311
3312out:
3313 ring_buffer_nest_end(buffer);
3314out_put:
3315 put_trace_buf();
3316
3317out_nobuffer:
3318 preempt_enable_notrace();
3319 unpause_graph_tracing();
3320
3321 return len;
3322}
3323EXPORT_SYMBOL_GPL(trace_vbprintk);
3324
3325__printf(3, 0)
3326static int
3327__trace_array_vprintk(struct trace_buffer *buffer,
3328 unsigned long ip, const char *fmt, va_list args)
3329{
3330 struct ring_buffer_event *event;
3331 int len = 0, size;
3332 struct print_entry *entry;
3333 unsigned int trace_ctx;
3334 char *tbuffer;
3335
3336 if (tracing_disabled)
3337 return 0;
3338
3339 /* Don't pollute graph traces with trace_vprintk internals */
3340 pause_graph_tracing();
3341
3342 trace_ctx = tracing_gen_ctx();
3343 preempt_disable_notrace();
3344
3345
3346 tbuffer = get_trace_buf();
3347 if (!tbuffer) {
3348 len = 0;
3349 goto out_nobuffer;
3350 }
3351
3352 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3353
3354 size = sizeof(*entry) + len + 1;
3355 ring_buffer_nest_start(buffer);
3356 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3357 trace_ctx);
3358 if (!event)
3359 goto out;
3360 entry = ring_buffer_event_data(event);
3361 entry->ip = ip;
3362
3363 memcpy(&entry->buf, tbuffer, len + 1);
3364 __buffer_unlock_commit(buffer, event);
3365 ftrace_trace_stack(printk_trace, buffer, trace_ctx, 6, NULL);
3366
3367out:
3368 ring_buffer_nest_end(buffer);
3369 put_trace_buf();
3370
3371out_nobuffer:
3372 preempt_enable_notrace();
3373 unpause_graph_tracing();
3374
3375 return len;
3376}
3377
3378__printf(3, 0)
3379int trace_array_vprintk(struct trace_array *tr,
3380 unsigned long ip, const char *fmt, va_list args)
3381{
3382 if (tracing_selftest_running && tr == &global_trace)
3383 return 0;
3384
3385 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3386}
3387
3388/**
3389 * trace_array_printk - Print a message to a specific instance
3390 * @tr: The instance trace_array descriptor
3391 * @ip: The instruction pointer that this is called from.
3392 * @fmt: The format to print (printf format)
3393 *
3394 * If a subsystem sets up its own instance, they have the right to
3395 * printk strings into their tracing instance buffer using this
3396 * function. Note, this function will not write into the top level
3397 * buffer (use trace_printk() for that), as writing into the top level
3398 * buffer should only have events that can be individually disabled.
3399 * trace_printk() is only used for debugging a kernel, and should not
3400 * be ever incorporated in normal use.
3401 *
3402 * trace_array_printk() can be used, as it will not add noise to the
3403 * top level tracing buffer.
3404 *
3405 * Note, trace_array_init_printk() must be called on @tr before this
3406 * can be used.
3407 */
3408__printf(3, 0)
3409int trace_array_printk(struct trace_array *tr,
3410 unsigned long ip, const char *fmt, ...)
3411{
3412 int ret;
3413 va_list ap;
3414
3415 if (!tr)
3416 return -ENOENT;
3417
3418 /* This is only allowed for created instances */
3419 if (tr == &global_trace)
3420 return 0;
3421
3422 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3423 return 0;
3424
3425 va_start(ap, fmt);
3426 ret = trace_array_vprintk(tr, ip, fmt, ap);
3427 va_end(ap);
3428 return ret;
3429}
3430EXPORT_SYMBOL_GPL(trace_array_printk);
3431
3432/**
3433 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3434 * @tr: The trace array to initialize the buffers for
3435 *
3436 * As trace_array_printk() only writes into instances, they are OK to
3437 * have in the kernel (unlike trace_printk()). This needs to be called
3438 * before trace_array_printk() can be used on a trace_array.
3439 */
3440int trace_array_init_printk(struct trace_array *tr)
3441{
3442 if (!tr)
3443 return -ENOENT;
3444
3445 /* This is only allowed for created instances */
3446 if (tr == &global_trace)
3447 return -EINVAL;
3448
3449 return alloc_percpu_trace_buffer();
3450}
3451EXPORT_SYMBOL_GPL(trace_array_init_printk);
3452
3453__printf(3, 4)
3454int trace_array_printk_buf(struct trace_buffer *buffer,
3455 unsigned long ip, const char *fmt, ...)
3456{
3457 int ret;
3458 va_list ap;
3459
3460 if (!(printk_trace->trace_flags & TRACE_ITER_PRINTK))
3461 return 0;
3462
3463 va_start(ap, fmt);
3464 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3465 va_end(ap);
3466 return ret;
3467}
3468
3469__printf(2, 0)
3470int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3471{
3472 return trace_array_vprintk(printk_trace, ip, fmt, args);
3473}
3474EXPORT_SYMBOL_GPL(trace_vprintk);
3475
3476static void trace_iterator_increment(struct trace_iterator *iter)
3477{
3478 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3479
3480 iter->idx++;
3481 if (buf_iter)
3482 ring_buffer_iter_advance(buf_iter);
3483}
3484
3485static struct trace_entry *
3486peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3487 unsigned long *lost_events)
3488{
3489 struct ring_buffer_event *event;
3490 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3491
3492 if (buf_iter) {
3493 event = ring_buffer_iter_peek(buf_iter, ts);
3494 if (lost_events)
3495 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3496 (unsigned long)-1 : 0;
3497 } else {
3498 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3499 lost_events);
3500 }
3501
3502 if (event) {
3503 iter->ent_size = ring_buffer_event_length(event);
3504 return ring_buffer_event_data(event);
3505 }
3506 iter->ent_size = 0;
3507 return NULL;
3508}
3509
3510static struct trace_entry *
3511__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3512 unsigned long *missing_events, u64 *ent_ts)
3513{
3514 struct trace_buffer *buffer = iter->array_buffer->buffer;
3515 struct trace_entry *ent, *next = NULL;
3516 unsigned long lost_events = 0, next_lost = 0;
3517 int cpu_file = iter->cpu_file;
3518 u64 next_ts = 0, ts;
3519 int next_cpu = -1;
3520 int next_size = 0;
3521 int cpu;
3522
3523 /*
3524 * If we are in a per_cpu trace file, don't bother by iterating over
3525 * all cpu and peek directly.
3526 */
3527 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3528 if (ring_buffer_empty_cpu(buffer, cpu_file))
3529 return NULL;
3530 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3531 if (ent_cpu)
3532 *ent_cpu = cpu_file;
3533
3534 return ent;
3535 }
3536
3537 for_each_tracing_cpu(cpu) {
3538
3539 if (ring_buffer_empty_cpu(buffer, cpu))
3540 continue;
3541
3542 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3543
3544 /*
3545 * Pick the entry with the smallest timestamp:
3546 */
3547 if (ent && (!next || ts < next_ts)) {
3548 next = ent;
3549 next_cpu = cpu;
3550 next_ts = ts;
3551 next_lost = lost_events;
3552 next_size = iter->ent_size;
3553 }
3554 }
3555
3556 iter->ent_size = next_size;
3557
3558 if (ent_cpu)
3559 *ent_cpu = next_cpu;
3560
3561 if (ent_ts)
3562 *ent_ts = next_ts;
3563
3564 if (missing_events)
3565 *missing_events = next_lost;
3566
3567 return next;
3568}
3569
3570#define STATIC_FMT_BUF_SIZE 128
3571static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3572
3573char *trace_iter_expand_format(struct trace_iterator *iter)
3574{
3575 char *tmp;
3576
3577 /*
3578 * iter->tr is NULL when used with tp_printk, which makes
3579 * this get called where it is not safe to call krealloc().
3580 */
3581 if (!iter->tr || iter->fmt == static_fmt_buf)
3582 return NULL;
3583
3584 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3585 GFP_KERNEL);
3586 if (tmp) {
3587 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3588 iter->fmt = tmp;
3589 }
3590
3591 return tmp;
3592}
3593
3594/* Returns true if the string is safe to dereference from an event */
3595static bool trace_safe_str(struct trace_iterator *iter, const char *str)
3596{
3597 unsigned long addr = (unsigned long)str;
3598 struct trace_event *trace_event;
3599 struct trace_event_call *event;
3600
3601 /* OK if part of the event data */
3602 if ((addr >= (unsigned long)iter->ent) &&
3603 (addr < (unsigned long)iter->ent + iter->ent_size))
3604 return true;
3605
3606 /* OK if part of the temp seq buffer */
3607 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3608 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3609 return true;
3610
3611 /* Core rodata can not be freed */
3612 if (is_kernel_rodata(addr))
3613 return true;
3614
3615 if (trace_is_tracepoint_string(str))
3616 return true;
3617
3618 /*
3619 * Now this could be a module event, referencing core module
3620 * data, which is OK.
3621 */
3622 if (!iter->ent)
3623 return false;
3624
3625 trace_event = ftrace_find_event(iter->ent->type);
3626 if (!trace_event)
3627 return false;
3628
3629 event = container_of(trace_event, struct trace_event_call, event);
3630 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3631 return false;
3632
3633 /* Would rather have rodata, but this will suffice */
3634 if (within_module_core(addr, event->module))
3635 return true;
3636
3637 return false;
3638}
3639
3640/**
3641 * ignore_event - Check dereferenced fields while writing to the seq buffer
3642 * @iter: The iterator that holds the seq buffer and the event being printed
3643 *
3644 * At boot up, test_event_printk() will flag any event that dereferences
3645 * a string with "%s" that does exist in the ring buffer. It may still
3646 * be valid, as the string may point to a static string in the kernel
3647 * rodata that never gets freed. But if the string pointer is pointing
3648 * to something that was allocated, there's a chance that it can be freed
3649 * by the time the user reads the trace. This would cause a bad memory
3650 * access by the kernel and possibly crash the system.
3651 *
3652 * This function will check if the event has any fields flagged as needing
3653 * to be checked at runtime and perform those checks.
3654 *
3655 * If it is found that a field is unsafe, it will write into the @iter->seq
3656 * a message stating what was found to be unsafe.
3657 *
3658 * @return: true if the event is unsafe and should be ignored,
3659 * false otherwise.
3660 */
3661bool ignore_event(struct trace_iterator *iter)
3662{
3663 struct ftrace_event_field *field;
3664 struct trace_event *trace_event;
3665 struct trace_event_call *event;
3666 struct list_head *head;
3667 struct trace_seq *seq;
3668 const void *ptr;
3669
3670 trace_event = ftrace_find_event(iter->ent->type);
3671
3672 seq = &iter->seq;
3673
3674 if (!trace_event) {
3675 trace_seq_printf(seq, "EVENT ID %d NOT FOUND?\n", iter->ent->type);
3676 return true;
3677 }
3678
3679 event = container_of(trace_event, struct trace_event_call, event);
3680 if (!(event->flags & TRACE_EVENT_FL_TEST_STR))
3681 return false;
3682
3683 head = trace_get_fields(event);
3684 if (!head) {
3685 trace_seq_printf(seq, "FIELDS FOR EVENT '%s' NOT FOUND?\n",
3686 trace_event_name(event));
3687 return true;
3688 }
3689
3690 /* Offsets are from the iter->ent that points to the raw event */
3691 ptr = iter->ent;
3692
3693 list_for_each_entry(field, head, link) {
3694 const char *str;
3695 bool good;
3696
3697 if (!field->needs_test)
3698 continue;
3699
3700 str = *(const char **)(ptr + field->offset);
3701
3702 good = trace_safe_str(iter, str);
3703
3704 /*
3705 * If you hit this warning, it is likely that the
3706 * trace event in question used %s on a string that
3707 * was saved at the time of the event, but may not be
3708 * around when the trace is read. Use __string(),
3709 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3710 * instead. See samples/trace_events/trace-events-sample.h
3711 * for reference.
3712 */
3713 if (WARN_ONCE(!good, "event '%s' has unsafe pointer field '%s'",
3714 trace_event_name(event), field->name)) {
3715 trace_seq_printf(seq, "EVENT %s: HAS UNSAFE POINTER FIELD '%s'\n",
3716 trace_event_name(event), field->name);
3717 return true;
3718 }
3719 }
3720 return false;
3721}
3722
3723const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3724{
3725 const char *p, *new_fmt;
3726 char *q;
3727
3728 if (WARN_ON_ONCE(!fmt))
3729 return fmt;
3730
3731 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3732 return fmt;
3733
3734 p = fmt;
3735 new_fmt = q = iter->fmt;
3736 while (*p) {
3737 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3738 if (!trace_iter_expand_format(iter))
3739 return fmt;
3740
3741 q += iter->fmt - new_fmt;
3742 new_fmt = iter->fmt;
3743 }
3744
3745 *q++ = *p++;
3746
3747 /* Replace %p with %px */
3748 if (p[-1] == '%') {
3749 if (p[0] == '%') {
3750 *q++ = *p++;
3751 } else if (p[0] == 'p' && !isalnum(p[1])) {
3752 *q++ = *p++;
3753 *q++ = 'x';
3754 }
3755 }
3756 }
3757 *q = '\0';
3758
3759 return new_fmt;
3760}
3761
3762#define STATIC_TEMP_BUF_SIZE 128
3763static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3764
3765/* Find the next real entry, without updating the iterator itself */
3766struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3767 int *ent_cpu, u64 *ent_ts)
3768{
3769 /* __find_next_entry will reset ent_size */
3770 int ent_size = iter->ent_size;
3771 struct trace_entry *entry;
3772
3773 /*
3774 * If called from ftrace_dump(), then the iter->temp buffer
3775 * will be the static_temp_buf and not created from kmalloc.
3776 * If the entry size is greater than the buffer, we can
3777 * not save it. Just return NULL in that case. This is only
3778 * used to add markers when two consecutive events' time
3779 * stamps have a large delta. See trace_print_lat_context()
3780 */
3781 if (iter->temp == static_temp_buf &&
3782 STATIC_TEMP_BUF_SIZE < ent_size)
3783 return NULL;
3784
3785 /*
3786 * The __find_next_entry() may call peek_next_entry(), which may
3787 * call ring_buffer_peek() that may make the contents of iter->ent
3788 * undefined. Need to copy iter->ent now.
3789 */
3790 if (iter->ent && iter->ent != iter->temp) {
3791 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3792 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3793 void *temp;
3794 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3795 if (!temp)
3796 return NULL;
3797 kfree(iter->temp);
3798 iter->temp = temp;
3799 iter->temp_size = iter->ent_size;
3800 }
3801 memcpy(iter->temp, iter->ent, iter->ent_size);
3802 iter->ent = iter->temp;
3803 }
3804 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3805 /* Put back the original ent_size */
3806 iter->ent_size = ent_size;
3807
3808 return entry;
3809}
3810
3811/* Find the next real entry, and increment the iterator to the next entry */
3812void *trace_find_next_entry_inc(struct trace_iterator *iter)
3813{
3814 iter->ent = __find_next_entry(iter, &iter->cpu,
3815 &iter->lost_events, &iter->ts);
3816
3817 if (iter->ent)
3818 trace_iterator_increment(iter);
3819
3820 return iter->ent ? iter : NULL;
3821}
3822
3823static void trace_consume(struct trace_iterator *iter)
3824{
3825 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3826 &iter->lost_events);
3827}
3828
3829static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3830{
3831 struct trace_iterator *iter = m->private;
3832 int i = (int)*pos;
3833 void *ent;
3834
3835 WARN_ON_ONCE(iter->leftover);
3836
3837 (*pos)++;
3838
3839 /* can't go backwards */
3840 if (iter->idx > i)
3841 return NULL;
3842
3843 if (iter->idx < 0)
3844 ent = trace_find_next_entry_inc(iter);
3845 else
3846 ent = iter;
3847
3848 while (ent && iter->idx < i)
3849 ent = trace_find_next_entry_inc(iter);
3850
3851 iter->pos = *pos;
3852
3853 return ent;
3854}
3855
3856void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3857{
3858 struct ring_buffer_iter *buf_iter;
3859 unsigned long entries = 0;
3860 u64 ts;
3861
3862 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3863
3864 buf_iter = trace_buffer_iter(iter, cpu);
3865 if (!buf_iter)
3866 return;
3867
3868 ring_buffer_iter_reset(buf_iter);
3869
3870 /*
3871 * We could have the case with the max latency tracers
3872 * that a reset never took place on a cpu. This is evident
3873 * by the timestamp being before the start of the buffer.
3874 */
3875 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3876 if (ts >= iter->array_buffer->time_start)
3877 break;
3878 entries++;
3879 ring_buffer_iter_advance(buf_iter);
3880 /* This could be a big loop */
3881 cond_resched();
3882 }
3883
3884 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3885}
3886
3887/*
3888 * The current tracer is copied to avoid a global locking
3889 * all around.
3890 */
3891static void *s_start(struct seq_file *m, loff_t *pos)
3892{
3893 struct trace_iterator *iter = m->private;
3894 struct trace_array *tr = iter->tr;
3895 int cpu_file = iter->cpu_file;
3896 void *p = NULL;
3897 loff_t l = 0;
3898 int cpu;
3899
3900 mutex_lock(&trace_types_lock);
3901 if (unlikely(tr->current_trace != iter->trace)) {
3902 /* Close iter->trace before switching to the new current tracer */
3903 if (iter->trace->close)
3904 iter->trace->close(iter);
3905 iter->trace = tr->current_trace;
3906 /* Reopen the new current tracer */
3907 if (iter->trace->open)
3908 iter->trace->open(iter);
3909 }
3910 mutex_unlock(&trace_types_lock);
3911
3912#ifdef CONFIG_TRACER_MAX_TRACE
3913 if (iter->snapshot && iter->trace->use_max_tr)
3914 return ERR_PTR(-EBUSY);
3915#endif
3916
3917 if (*pos != iter->pos) {
3918 iter->ent = NULL;
3919 iter->cpu = 0;
3920 iter->idx = -1;
3921
3922 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3923 for_each_tracing_cpu(cpu)
3924 tracing_iter_reset(iter, cpu);
3925 } else
3926 tracing_iter_reset(iter, cpu_file);
3927
3928 iter->leftover = 0;
3929 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3930 ;
3931
3932 } else {
3933 /*
3934 * If we overflowed the seq_file before, then we want
3935 * to just reuse the trace_seq buffer again.
3936 */
3937 if (iter->leftover)
3938 p = iter;
3939 else {
3940 l = *pos - 1;
3941 p = s_next(m, p, &l);
3942 }
3943 }
3944
3945 trace_event_read_lock();
3946 trace_access_lock(cpu_file);
3947 return p;
3948}
3949
3950static void s_stop(struct seq_file *m, void *p)
3951{
3952 struct trace_iterator *iter = m->private;
3953
3954#ifdef CONFIG_TRACER_MAX_TRACE
3955 if (iter->snapshot && iter->trace->use_max_tr)
3956 return;
3957#endif
3958
3959 trace_access_unlock(iter->cpu_file);
3960 trace_event_read_unlock();
3961}
3962
3963static void
3964get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3965 unsigned long *entries, int cpu)
3966{
3967 unsigned long count;
3968
3969 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3970 /*
3971 * If this buffer has skipped entries, then we hold all
3972 * entries for the trace and we need to ignore the
3973 * ones before the time stamp.
3974 */
3975 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3976 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3977 /* total is the same as the entries */
3978 *total = count;
3979 } else
3980 *total = count +
3981 ring_buffer_overrun_cpu(buf->buffer, cpu);
3982 *entries = count;
3983}
3984
3985static void
3986get_total_entries(struct array_buffer *buf,
3987 unsigned long *total, unsigned long *entries)
3988{
3989 unsigned long t, e;
3990 int cpu;
3991
3992 *total = 0;
3993 *entries = 0;
3994
3995 for_each_tracing_cpu(cpu) {
3996 get_total_entries_cpu(buf, &t, &e, cpu);
3997 *total += t;
3998 *entries += e;
3999 }
4000}
4001
4002unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4003{
4004 unsigned long total, entries;
4005
4006 if (!tr)
4007 tr = &global_trace;
4008
4009 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4010
4011 return entries;
4012}
4013
4014unsigned long trace_total_entries(struct trace_array *tr)
4015{
4016 unsigned long total, entries;
4017
4018 if (!tr)
4019 tr = &global_trace;
4020
4021 get_total_entries(&tr->array_buffer, &total, &entries);
4022
4023 return entries;
4024}
4025
4026static void print_lat_help_header(struct seq_file *m)
4027{
4028 seq_puts(m, "# _------=> CPU# \n"
4029 "# / _-----=> irqs-off/BH-disabled\n"
4030 "# | / _----=> need-resched \n"
4031 "# || / _---=> hardirq/softirq \n"
4032 "# ||| / _--=> preempt-depth \n"
4033 "# |||| / _-=> migrate-disable \n"
4034 "# ||||| / delay \n"
4035 "# cmd pid |||||| time | caller \n"
4036 "# \\ / |||||| \\ | / \n");
4037}
4038
4039static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4040{
4041 unsigned long total;
4042 unsigned long entries;
4043
4044 get_total_entries(buf, &total, &entries);
4045 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4046 entries, total, num_online_cpus());
4047 seq_puts(m, "#\n");
4048}
4049
4050static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4051 unsigned int flags)
4052{
4053 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4054
4055 print_event_info(buf, m);
4056
4057 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4058 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4059}
4060
4061static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4062 unsigned int flags)
4063{
4064 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4065 static const char space[] = " ";
4066 int prec = tgid ? 12 : 2;
4067
4068 print_event_info(buf, m);
4069
4070 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4071 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4072 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4073 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4074 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4075 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4076 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4077 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4078}
4079
4080void
4081print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4082{
4083 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4084 struct array_buffer *buf = iter->array_buffer;
4085 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4086 struct tracer *type = iter->trace;
4087 unsigned long entries;
4088 unsigned long total;
4089 const char *name = type->name;
4090
4091 get_total_entries(buf, &total, &entries);
4092
4093 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4094 name, init_utsname()->release);
4095 seq_puts(m, "# -----------------------------------"
4096 "---------------------------------\n");
4097 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4098 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4099 nsecs_to_usecs(data->saved_latency),
4100 entries,
4101 total,
4102 buf->cpu,
4103 preempt_model_none() ? "server" :
4104 preempt_model_voluntary() ? "desktop" :
4105 preempt_model_full() ? "preempt" :
4106 preempt_model_lazy() ? "lazy" :
4107 preempt_model_rt() ? "preempt_rt" :
4108 "unknown",
4109 /* These are reserved for later use */
4110 0, 0, 0, 0);
4111#ifdef CONFIG_SMP
4112 seq_printf(m, " #P:%d)\n", num_online_cpus());
4113#else
4114 seq_puts(m, ")\n");
4115#endif
4116 seq_puts(m, "# -----------------\n");
4117 seq_printf(m, "# | task: %.16s-%d "
4118 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4119 data->comm, data->pid,
4120 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4121 data->policy, data->rt_priority);
4122 seq_puts(m, "# -----------------\n");
4123
4124 if (data->critical_start) {
4125 seq_puts(m, "# => started at: ");
4126 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4127 trace_print_seq(m, &iter->seq);
4128 seq_puts(m, "\n# => ended at: ");
4129 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4130 trace_print_seq(m, &iter->seq);
4131 seq_puts(m, "\n#\n");
4132 }
4133
4134 seq_puts(m, "#\n");
4135}
4136
4137static void test_cpu_buff_start(struct trace_iterator *iter)
4138{
4139 struct trace_seq *s = &iter->seq;
4140 struct trace_array *tr = iter->tr;
4141
4142 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4143 return;
4144
4145 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4146 return;
4147
4148 if (cpumask_available(iter->started) &&
4149 cpumask_test_cpu(iter->cpu, iter->started))
4150 return;
4151
4152 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4153 return;
4154
4155 if (cpumask_available(iter->started))
4156 cpumask_set_cpu(iter->cpu, iter->started);
4157
4158 /* Don't print started cpu buffer for the first entry of the trace */
4159 if (iter->idx > 1)
4160 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4161 iter->cpu);
4162}
4163
4164static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4165{
4166 struct trace_array *tr = iter->tr;
4167 struct trace_seq *s = &iter->seq;
4168 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4169 struct trace_entry *entry;
4170 struct trace_event *event;
4171
4172 entry = iter->ent;
4173
4174 test_cpu_buff_start(iter);
4175
4176 event = ftrace_find_event(entry->type);
4177
4178 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4179 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4180 trace_print_lat_context(iter);
4181 else
4182 trace_print_context(iter);
4183 }
4184
4185 if (trace_seq_has_overflowed(s))
4186 return TRACE_TYPE_PARTIAL_LINE;
4187
4188 if (event) {
4189 if (tr->trace_flags & TRACE_ITER_FIELDS)
4190 return print_event_fields(iter, event);
4191 /*
4192 * For TRACE_EVENT() events, the print_fmt is not
4193 * safe to use if the array has delta offsets
4194 * Force printing via the fields.
4195 */
4196 if ((tr->text_delta || tr->data_delta) &&
4197 event->type > __TRACE_LAST_TYPE)
4198 return print_event_fields(iter, event);
4199
4200 return event->funcs->trace(iter, sym_flags, event);
4201 }
4202
4203 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4204
4205 return trace_handle_return(s);
4206}
4207
4208static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4209{
4210 struct trace_array *tr = iter->tr;
4211 struct trace_seq *s = &iter->seq;
4212 struct trace_entry *entry;
4213 struct trace_event *event;
4214
4215 entry = iter->ent;
4216
4217 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4218 trace_seq_printf(s, "%d %d %llu ",
4219 entry->pid, iter->cpu, iter->ts);
4220
4221 if (trace_seq_has_overflowed(s))
4222 return TRACE_TYPE_PARTIAL_LINE;
4223
4224 event = ftrace_find_event(entry->type);
4225 if (event)
4226 return event->funcs->raw(iter, 0, event);
4227
4228 trace_seq_printf(s, "%d ?\n", entry->type);
4229
4230 return trace_handle_return(s);
4231}
4232
4233static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4234{
4235 struct trace_array *tr = iter->tr;
4236 struct trace_seq *s = &iter->seq;
4237 unsigned char newline = '\n';
4238 struct trace_entry *entry;
4239 struct trace_event *event;
4240
4241 entry = iter->ent;
4242
4243 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4244 SEQ_PUT_HEX_FIELD(s, entry->pid);
4245 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4246 SEQ_PUT_HEX_FIELD(s, iter->ts);
4247 if (trace_seq_has_overflowed(s))
4248 return TRACE_TYPE_PARTIAL_LINE;
4249 }
4250
4251 event = ftrace_find_event(entry->type);
4252 if (event) {
4253 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4254 if (ret != TRACE_TYPE_HANDLED)
4255 return ret;
4256 }
4257
4258 SEQ_PUT_FIELD(s, newline);
4259
4260 return trace_handle_return(s);
4261}
4262
4263static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4264{
4265 struct trace_array *tr = iter->tr;
4266 struct trace_seq *s = &iter->seq;
4267 struct trace_entry *entry;
4268 struct trace_event *event;
4269
4270 entry = iter->ent;
4271
4272 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4273 SEQ_PUT_FIELD(s, entry->pid);
4274 SEQ_PUT_FIELD(s, iter->cpu);
4275 SEQ_PUT_FIELD(s, iter->ts);
4276 if (trace_seq_has_overflowed(s))
4277 return TRACE_TYPE_PARTIAL_LINE;
4278 }
4279
4280 event = ftrace_find_event(entry->type);
4281 return event ? event->funcs->binary(iter, 0, event) :
4282 TRACE_TYPE_HANDLED;
4283}
4284
4285int trace_empty(struct trace_iterator *iter)
4286{
4287 struct ring_buffer_iter *buf_iter;
4288 int cpu;
4289
4290 /* If we are looking at one CPU buffer, only check that one */
4291 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4292 cpu = iter->cpu_file;
4293 buf_iter = trace_buffer_iter(iter, cpu);
4294 if (buf_iter) {
4295 if (!ring_buffer_iter_empty(buf_iter))
4296 return 0;
4297 } else {
4298 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4299 return 0;
4300 }
4301 return 1;
4302 }
4303
4304 for_each_tracing_cpu(cpu) {
4305 buf_iter = trace_buffer_iter(iter, cpu);
4306 if (buf_iter) {
4307 if (!ring_buffer_iter_empty(buf_iter))
4308 return 0;
4309 } else {
4310 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4311 return 0;
4312 }
4313 }
4314
4315 return 1;
4316}
4317
4318/* Called with trace_event_read_lock() held. */
4319enum print_line_t print_trace_line(struct trace_iterator *iter)
4320{
4321 struct trace_array *tr = iter->tr;
4322 unsigned long trace_flags = tr->trace_flags;
4323 enum print_line_t ret;
4324
4325 if (iter->lost_events) {
4326 if (iter->lost_events == (unsigned long)-1)
4327 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4328 iter->cpu);
4329 else
4330 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4331 iter->cpu, iter->lost_events);
4332 if (trace_seq_has_overflowed(&iter->seq))
4333 return TRACE_TYPE_PARTIAL_LINE;
4334 }
4335
4336 if (iter->trace && iter->trace->print_line) {
4337 ret = iter->trace->print_line(iter);
4338 if (ret != TRACE_TYPE_UNHANDLED)
4339 return ret;
4340 }
4341
4342 if (iter->ent->type == TRACE_BPUTS &&
4343 trace_flags & TRACE_ITER_PRINTK &&
4344 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4345 return trace_print_bputs_msg_only(iter);
4346
4347 if (iter->ent->type == TRACE_BPRINT &&
4348 trace_flags & TRACE_ITER_PRINTK &&
4349 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4350 return trace_print_bprintk_msg_only(iter);
4351
4352 if (iter->ent->type == TRACE_PRINT &&
4353 trace_flags & TRACE_ITER_PRINTK &&
4354 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4355 return trace_print_printk_msg_only(iter);
4356
4357 if (trace_flags & TRACE_ITER_BIN)
4358 return print_bin_fmt(iter);
4359
4360 if (trace_flags & TRACE_ITER_HEX)
4361 return print_hex_fmt(iter);
4362
4363 if (trace_flags & TRACE_ITER_RAW)
4364 return print_raw_fmt(iter);
4365
4366 return print_trace_fmt(iter);
4367}
4368
4369void trace_latency_header(struct seq_file *m)
4370{
4371 struct trace_iterator *iter = m->private;
4372 struct trace_array *tr = iter->tr;
4373
4374 /* print nothing if the buffers are empty */
4375 if (trace_empty(iter))
4376 return;
4377
4378 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4379 print_trace_header(m, iter);
4380
4381 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4382 print_lat_help_header(m);
4383}
4384
4385void trace_default_header(struct seq_file *m)
4386{
4387 struct trace_iterator *iter = m->private;
4388 struct trace_array *tr = iter->tr;
4389 unsigned long trace_flags = tr->trace_flags;
4390
4391 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4392 return;
4393
4394 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4395 /* print nothing if the buffers are empty */
4396 if (trace_empty(iter))
4397 return;
4398 print_trace_header(m, iter);
4399 if (!(trace_flags & TRACE_ITER_VERBOSE))
4400 print_lat_help_header(m);
4401 } else {
4402 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4403 if (trace_flags & TRACE_ITER_IRQ_INFO)
4404 print_func_help_header_irq(iter->array_buffer,
4405 m, trace_flags);
4406 else
4407 print_func_help_header(iter->array_buffer, m,
4408 trace_flags);
4409 }
4410 }
4411}
4412
4413static void test_ftrace_alive(struct seq_file *m)
4414{
4415 if (!ftrace_is_dead())
4416 return;
4417 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4418 "# MAY BE MISSING FUNCTION EVENTS\n");
4419}
4420
4421#ifdef CONFIG_TRACER_MAX_TRACE
4422static void show_snapshot_main_help(struct seq_file *m)
4423{
4424 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4425 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4426 "# Takes a snapshot of the main buffer.\n"
4427 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4428 "# (Doesn't have to be '2' works with any number that\n"
4429 "# is not a '0' or '1')\n");
4430}
4431
4432static void show_snapshot_percpu_help(struct seq_file *m)
4433{
4434 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4435#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4436 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4437 "# Takes a snapshot of the main buffer for this cpu.\n");
4438#else
4439 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4440 "# Must use main snapshot file to allocate.\n");
4441#endif
4442 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4443 "# (Doesn't have to be '2' works with any number that\n"
4444 "# is not a '0' or '1')\n");
4445}
4446
4447static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4448{
4449 if (iter->tr->allocated_snapshot)
4450 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4451 else
4452 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4453
4454 seq_puts(m, "# Snapshot commands:\n");
4455 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4456 show_snapshot_main_help(m);
4457 else
4458 show_snapshot_percpu_help(m);
4459}
4460#else
4461/* Should never be called */
4462static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4463#endif
4464
4465static int s_show(struct seq_file *m, void *v)
4466{
4467 struct trace_iterator *iter = v;
4468 int ret;
4469
4470 if (iter->ent == NULL) {
4471 if (iter->tr) {
4472 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4473 seq_puts(m, "#\n");
4474 test_ftrace_alive(m);
4475 }
4476 if (iter->snapshot && trace_empty(iter))
4477 print_snapshot_help(m, iter);
4478 else if (iter->trace && iter->trace->print_header)
4479 iter->trace->print_header(m);
4480 else
4481 trace_default_header(m);
4482
4483 } else if (iter->leftover) {
4484 /*
4485 * If we filled the seq_file buffer earlier, we
4486 * want to just show it now.
4487 */
4488 ret = trace_print_seq(m, &iter->seq);
4489
4490 /* ret should this time be zero, but you never know */
4491 iter->leftover = ret;
4492
4493 } else {
4494 ret = print_trace_line(iter);
4495 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4496 iter->seq.full = 0;
4497 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4498 }
4499 ret = trace_print_seq(m, &iter->seq);
4500 /*
4501 * If we overflow the seq_file buffer, then it will
4502 * ask us for this data again at start up.
4503 * Use that instead.
4504 * ret is 0 if seq_file write succeeded.
4505 * -1 otherwise.
4506 */
4507 iter->leftover = ret;
4508 }
4509
4510 return 0;
4511}
4512
4513/*
4514 * Should be used after trace_array_get(), trace_types_lock
4515 * ensures that i_cdev was already initialized.
4516 */
4517static inline int tracing_get_cpu(struct inode *inode)
4518{
4519 if (inode->i_cdev) /* See trace_create_cpu_file() */
4520 return (long)inode->i_cdev - 1;
4521 return RING_BUFFER_ALL_CPUS;
4522}
4523
4524static const struct seq_operations tracer_seq_ops = {
4525 .start = s_start,
4526 .next = s_next,
4527 .stop = s_stop,
4528 .show = s_show,
4529};
4530
4531/*
4532 * Note, as iter itself can be allocated and freed in different
4533 * ways, this function is only used to free its content, and not
4534 * the iterator itself. The only requirement to all the allocations
4535 * is that it must zero all fields (kzalloc), as freeing works with
4536 * ethier allocated content or NULL.
4537 */
4538static void free_trace_iter_content(struct trace_iterator *iter)
4539{
4540 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4541 if (iter->fmt != static_fmt_buf)
4542 kfree(iter->fmt);
4543
4544 kfree(iter->temp);
4545 kfree(iter->buffer_iter);
4546 mutex_destroy(&iter->mutex);
4547 free_cpumask_var(iter->started);
4548}
4549
4550static struct trace_iterator *
4551__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4552{
4553 struct trace_array *tr = inode->i_private;
4554 struct trace_iterator *iter;
4555 int cpu;
4556
4557 if (tracing_disabled)
4558 return ERR_PTR(-ENODEV);
4559
4560 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4561 if (!iter)
4562 return ERR_PTR(-ENOMEM);
4563
4564 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4565 GFP_KERNEL);
4566 if (!iter->buffer_iter)
4567 goto release;
4568
4569 /*
4570 * trace_find_next_entry() may need to save off iter->ent.
4571 * It will place it into the iter->temp buffer. As most
4572 * events are less than 128, allocate a buffer of that size.
4573 * If one is greater, then trace_find_next_entry() will
4574 * allocate a new buffer to adjust for the bigger iter->ent.
4575 * It's not critical if it fails to get allocated here.
4576 */
4577 iter->temp = kmalloc(128, GFP_KERNEL);
4578 if (iter->temp)
4579 iter->temp_size = 128;
4580
4581 /*
4582 * trace_event_printf() may need to modify given format
4583 * string to replace %p with %px so that it shows real address
4584 * instead of hash value. However, that is only for the event
4585 * tracing, other tracer may not need. Defer the allocation
4586 * until it is needed.
4587 */
4588 iter->fmt = NULL;
4589 iter->fmt_size = 0;
4590
4591 mutex_lock(&trace_types_lock);
4592 iter->trace = tr->current_trace;
4593
4594 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4595 goto fail;
4596
4597 iter->tr = tr;
4598
4599#ifdef CONFIG_TRACER_MAX_TRACE
4600 /* Currently only the top directory has a snapshot */
4601 if (tr->current_trace->print_max || snapshot)
4602 iter->array_buffer = &tr->max_buffer;
4603 else
4604#endif
4605 iter->array_buffer = &tr->array_buffer;
4606 iter->snapshot = snapshot;
4607 iter->pos = -1;
4608 iter->cpu_file = tracing_get_cpu(inode);
4609 mutex_init(&iter->mutex);
4610
4611 /* Notify the tracer early; before we stop tracing. */
4612 if (iter->trace->open)
4613 iter->trace->open(iter);
4614
4615 /* Annotate start of buffers if we had overruns */
4616 if (ring_buffer_overruns(iter->array_buffer->buffer))
4617 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4618
4619 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4620 if (trace_clocks[tr->clock_id].in_ns)
4621 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4622
4623 /*
4624 * If pause-on-trace is enabled, then stop the trace while
4625 * dumping, unless this is the "snapshot" file
4626 */
4627 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4628 tracing_stop_tr(tr);
4629
4630 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4631 for_each_tracing_cpu(cpu) {
4632 iter->buffer_iter[cpu] =
4633 ring_buffer_read_prepare(iter->array_buffer->buffer,
4634 cpu, GFP_KERNEL);
4635 }
4636 ring_buffer_read_prepare_sync();
4637 for_each_tracing_cpu(cpu) {
4638 ring_buffer_read_start(iter->buffer_iter[cpu]);
4639 tracing_iter_reset(iter, cpu);
4640 }
4641 } else {
4642 cpu = iter->cpu_file;
4643 iter->buffer_iter[cpu] =
4644 ring_buffer_read_prepare(iter->array_buffer->buffer,
4645 cpu, GFP_KERNEL);
4646 ring_buffer_read_prepare_sync();
4647 ring_buffer_read_start(iter->buffer_iter[cpu]);
4648 tracing_iter_reset(iter, cpu);
4649 }
4650
4651 mutex_unlock(&trace_types_lock);
4652
4653 return iter;
4654
4655 fail:
4656 mutex_unlock(&trace_types_lock);
4657 free_trace_iter_content(iter);
4658release:
4659 seq_release_private(inode, file);
4660 return ERR_PTR(-ENOMEM);
4661}
4662
4663int tracing_open_generic(struct inode *inode, struct file *filp)
4664{
4665 int ret;
4666
4667 ret = tracing_check_open_get_tr(NULL);
4668 if (ret)
4669 return ret;
4670
4671 filp->private_data = inode->i_private;
4672 return 0;
4673}
4674
4675bool tracing_is_disabled(void)
4676{
4677 return (tracing_disabled) ? true: false;
4678}
4679
4680/*
4681 * Open and update trace_array ref count.
4682 * Must have the current trace_array passed to it.
4683 */
4684int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4685{
4686 struct trace_array *tr = inode->i_private;
4687 int ret;
4688
4689 ret = tracing_check_open_get_tr(tr);
4690 if (ret)
4691 return ret;
4692
4693 filp->private_data = inode->i_private;
4694
4695 return 0;
4696}
4697
4698/*
4699 * The private pointer of the inode is the trace_event_file.
4700 * Update the tr ref count associated to it.
4701 */
4702int tracing_open_file_tr(struct inode *inode, struct file *filp)
4703{
4704 struct trace_event_file *file = inode->i_private;
4705 int ret;
4706
4707 ret = tracing_check_open_get_tr(file->tr);
4708 if (ret)
4709 return ret;
4710
4711 mutex_lock(&event_mutex);
4712
4713 /* Fail if the file is marked for removal */
4714 if (file->flags & EVENT_FILE_FL_FREED) {
4715 trace_array_put(file->tr);
4716 ret = -ENODEV;
4717 } else {
4718 event_file_get(file);
4719 }
4720
4721 mutex_unlock(&event_mutex);
4722 if (ret)
4723 return ret;
4724
4725 filp->private_data = inode->i_private;
4726
4727 return 0;
4728}
4729
4730int tracing_release_file_tr(struct inode *inode, struct file *filp)
4731{
4732 struct trace_event_file *file = inode->i_private;
4733
4734 trace_array_put(file->tr);
4735 event_file_put(file);
4736
4737 return 0;
4738}
4739
4740int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
4741{
4742 tracing_release_file_tr(inode, filp);
4743 return single_release(inode, filp);
4744}
4745
4746static int tracing_mark_open(struct inode *inode, struct file *filp)
4747{
4748 stream_open(inode, filp);
4749 return tracing_open_generic_tr(inode, filp);
4750}
4751
4752static int tracing_release(struct inode *inode, struct file *file)
4753{
4754 struct trace_array *tr = inode->i_private;
4755 struct seq_file *m = file->private_data;
4756 struct trace_iterator *iter;
4757 int cpu;
4758
4759 if (!(file->f_mode & FMODE_READ)) {
4760 trace_array_put(tr);
4761 return 0;
4762 }
4763
4764 /* Writes do not use seq_file */
4765 iter = m->private;
4766 mutex_lock(&trace_types_lock);
4767
4768 for_each_tracing_cpu(cpu) {
4769 if (iter->buffer_iter[cpu])
4770 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4771 }
4772
4773 if (iter->trace && iter->trace->close)
4774 iter->trace->close(iter);
4775
4776 if (!iter->snapshot && tr->stop_count)
4777 /* reenable tracing if it was previously enabled */
4778 tracing_start_tr(tr);
4779
4780 __trace_array_put(tr);
4781
4782 mutex_unlock(&trace_types_lock);
4783
4784 free_trace_iter_content(iter);
4785 seq_release_private(inode, file);
4786
4787 return 0;
4788}
4789
4790int tracing_release_generic_tr(struct inode *inode, struct file *file)
4791{
4792 struct trace_array *tr = inode->i_private;
4793
4794 trace_array_put(tr);
4795 return 0;
4796}
4797
4798static int tracing_single_release_tr(struct inode *inode, struct file *file)
4799{
4800 struct trace_array *tr = inode->i_private;
4801
4802 trace_array_put(tr);
4803
4804 return single_release(inode, file);
4805}
4806
4807static int tracing_open(struct inode *inode, struct file *file)
4808{
4809 struct trace_array *tr = inode->i_private;
4810 struct trace_iterator *iter;
4811 int ret;
4812
4813 ret = tracing_check_open_get_tr(tr);
4814 if (ret)
4815 return ret;
4816
4817 /* If this file was open for write, then erase contents */
4818 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4819 int cpu = tracing_get_cpu(inode);
4820 struct array_buffer *trace_buf = &tr->array_buffer;
4821
4822#ifdef CONFIG_TRACER_MAX_TRACE
4823 if (tr->current_trace->print_max)
4824 trace_buf = &tr->max_buffer;
4825#endif
4826
4827 if (cpu == RING_BUFFER_ALL_CPUS)
4828 tracing_reset_online_cpus(trace_buf);
4829 else
4830 tracing_reset_cpu(trace_buf, cpu);
4831 }
4832
4833 if (file->f_mode & FMODE_READ) {
4834 iter = __tracing_open(inode, file, false);
4835 if (IS_ERR(iter))
4836 ret = PTR_ERR(iter);
4837 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4838 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4839 }
4840
4841 if (ret < 0)
4842 trace_array_put(tr);
4843
4844 return ret;
4845}
4846
4847/*
4848 * Some tracers are not suitable for instance buffers.
4849 * A tracer is always available for the global array (toplevel)
4850 * or if it explicitly states that it is.
4851 */
4852static bool
4853trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4854{
4855#ifdef CONFIG_TRACER_SNAPSHOT
4856 /* arrays with mapped buffer range do not have snapshots */
4857 if (tr->range_addr_start && t->use_max_tr)
4858 return false;
4859#endif
4860 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4861}
4862
4863/* Find the next tracer that this trace array may use */
4864static struct tracer *
4865get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4866{
4867 while (t && !trace_ok_for_array(t, tr))
4868 t = t->next;
4869
4870 return t;
4871}
4872
4873static void *
4874t_next(struct seq_file *m, void *v, loff_t *pos)
4875{
4876 struct trace_array *tr = m->private;
4877 struct tracer *t = v;
4878
4879 (*pos)++;
4880
4881 if (t)
4882 t = get_tracer_for_array(tr, t->next);
4883
4884 return t;
4885}
4886
4887static void *t_start(struct seq_file *m, loff_t *pos)
4888{
4889 struct trace_array *tr = m->private;
4890 struct tracer *t;
4891 loff_t l = 0;
4892
4893 mutex_lock(&trace_types_lock);
4894
4895 t = get_tracer_for_array(tr, trace_types);
4896 for (; t && l < *pos; t = t_next(m, t, &l))
4897 ;
4898
4899 return t;
4900}
4901
4902static void t_stop(struct seq_file *m, void *p)
4903{
4904 mutex_unlock(&trace_types_lock);
4905}
4906
4907static int t_show(struct seq_file *m, void *v)
4908{
4909 struct tracer *t = v;
4910
4911 if (!t)
4912 return 0;
4913
4914 seq_puts(m, t->name);
4915 if (t->next)
4916 seq_putc(m, ' ');
4917 else
4918 seq_putc(m, '\n');
4919
4920 return 0;
4921}
4922
4923static const struct seq_operations show_traces_seq_ops = {
4924 .start = t_start,
4925 .next = t_next,
4926 .stop = t_stop,
4927 .show = t_show,
4928};
4929
4930static int show_traces_open(struct inode *inode, struct file *file)
4931{
4932 struct trace_array *tr = inode->i_private;
4933 struct seq_file *m;
4934 int ret;
4935
4936 ret = tracing_check_open_get_tr(tr);
4937 if (ret)
4938 return ret;
4939
4940 ret = seq_open(file, &show_traces_seq_ops);
4941 if (ret) {
4942 trace_array_put(tr);
4943 return ret;
4944 }
4945
4946 m = file->private_data;
4947 m->private = tr;
4948
4949 return 0;
4950}
4951
4952static int tracing_seq_release(struct inode *inode, struct file *file)
4953{
4954 struct trace_array *tr = inode->i_private;
4955
4956 trace_array_put(tr);
4957 return seq_release(inode, file);
4958}
4959
4960static ssize_t
4961tracing_write_stub(struct file *filp, const char __user *ubuf,
4962 size_t count, loff_t *ppos)
4963{
4964 return count;
4965}
4966
4967loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4968{
4969 int ret;
4970
4971 if (file->f_mode & FMODE_READ)
4972 ret = seq_lseek(file, offset, whence);
4973 else
4974 file->f_pos = ret = 0;
4975
4976 return ret;
4977}
4978
4979static const struct file_operations tracing_fops = {
4980 .open = tracing_open,
4981 .read = seq_read,
4982 .read_iter = seq_read_iter,
4983 .splice_read = copy_splice_read,
4984 .write = tracing_write_stub,
4985 .llseek = tracing_lseek,
4986 .release = tracing_release,
4987};
4988
4989static const struct file_operations show_traces_fops = {
4990 .open = show_traces_open,
4991 .read = seq_read,
4992 .llseek = seq_lseek,
4993 .release = tracing_seq_release,
4994};
4995
4996static ssize_t
4997tracing_cpumask_read(struct file *filp, char __user *ubuf,
4998 size_t count, loff_t *ppos)
4999{
5000 struct trace_array *tr = file_inode(filp)->i_private;
5001 char *mask_str;
5002 int len;
5003
5004 len = snprintf(NULL, 0, "%*pb\n",
5005 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5006 mask_str = kmalloc(len, GFP_KERNEL);
5007 if (!mask_str)
5008 return -ENOMEM;
5009
5010 len = snprintf(mask_str, len, "%*pb\n",
5011 cpumask_pr_args(tr->tracing_cpumask));
5012 if (len >= count) {
5013 count = -EINVAL;
5014 goto out_err;
5015 }
5016 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5017
5018out_err:
5019 kfree(mask_str);
5020
5021 return count;
5022}
5023
5024int tracing_set_cpumask(struct trace_array *tr,
5025 cpumask_var_t tracing_cpumask_new)
5026{
5027 int cpu;
5028
5029 if (!tr)
5030 return -EINVAL;
5031
5032 local_irq_disable();
5033 arch_spin_lock(&tr->max_lock);
5034 for_each_tracing_cpu(cpu) {
5035 /*
5036 * Increase/decrease the disabled counter if we are
5037 * about to flip a bit in the cpumask:
5038 */
5039 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5040 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5041 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5042 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5043#ifdef CONFIG_TRACER_MAX_TRACE
5044 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5045#endif
5046 }
5047 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5048 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5049 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5050 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5051#ifdef CONFIG_TRACER_MAX_TRACE
5052 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5053#endif
5054 }
5055 }
5056 arch_spin_unlock(&tr->max_lock);
5057 local_irq_enable();
5058
5059 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5060
5061 return 0;
5062}
5063
5064static ssize_t
5065tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5066 size_t count, loff_t *ppos)
5067{
5068 struct trace_array *tr = file_inode(filp)->i_private;
5069 cpumask_var_t tracing_cpumask_new;
5070 int err;
5071
5072 if (count == 0 || count > KMALLOC_MAX_SIZE)
5073 return -EINVAL;
5074
5075 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5076 return -ENOMEM;
5077
5078 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5079 if (err)
5080 goto err_free;
5081
5082 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5083 if (err)
5084 goto err_free;
5085
5086 free_cpumask_var(tracing_cpumask_new);
5087
5088 return count;
5089
5090err_free:
5091 free_cpumask_var(tracing_cpumask_new);
5092
5093 return err;
5094}
5095
5096static const struct file_operations tracing_cpumask_fops = {
5097 .open = tracing_open_generic_tr,
5098 .read = tracing_cpumask_read,
5099 .write = tracing_cpumask_write,
5100 .release = tracing_release_generic_tr,
5101 .llseek = generic_file_llseek,
5102};
5103
5104static int tracing_trace_options_show(struct seq_file *m, void *v)
5105{
5106 struct tracer_opt *trace_opts;
5107 struct trace_array *tr = m->private;
5108 u32 tracer_flags;
5109 int i;
5110
5111 guard(mutex)(&trace_types_lock);
5112
5113 tracer_flags = tr->current_trace->flags->val;
5114 trace_opts = tr->current_trace->flags->opts;
5115
5116 for (i = 0; trace_options[i]; i++) {
5117 if (tr->trace_flags & (1 << i))
5118 seq_printf(m, "%s\n", trace_options[i]);
5119 else
5120 seq_printf(m, "no%s\n", trace_options[i]);
5121 }
5122
5123 for (i = 0; trace_opts[i].name; i++) {
5124 if (tracer_flags & trace_opts[i].bit)
5125 seq_printf(m, "%s\n", trace_opts[i].name);
5126 else
5127 seq_printf(m, "no%s\n", trace_opts[i].name);
5128 }
5129
5130 return 0;
5131}
5132
5133static int __set_tracer_option(struct trace_array *tr,
5134 struct tracer_flags *tracer_flags,
5135 struct tracer_opt *opts, int neg)
5136{
5137 struct tracer *trace = tracer_flags->trace;
5138 int ret;
5139
5140 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5141 if (ret)
5142 return ret;
5143
5144 if (neg)
5145 tracer_flags->val &= ~opts->bit;
5146 else
5147 tracer_flags->val |= opts->bit;
5148 return 0;
5149}
5150
5151/* Try to assign a tracer specific option */
5152static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5153{
5154 struct tracer *trace = tr->current_trace;
5155 struct tracer_flags *tracer_flags = trace->flags;
5156 struct tracer_opt *opts = NULL;
5157 int i;
5158
5159 for (i = 0; tracer_flags->opts[i].name; i++) {
5160 opts = &tracer_flags->opts[i];
5161
5162 if (strcmp(cmp, opts->name) == 0)
5163 return __set_tracer_option(tr, trace->flags, opts, neg);
5164 }
5165
5166 return -EINVAL;
5167}
5168
5169/* Some tracers require overwrite to stay enabled */
5170int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5171{
5172 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5173 return -1;
5174
5175 return 0;
5176}
5177
5178int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5179{
5180 if ((mask == TRACE_ITER_RECORD_TGID) ||
5181 (mask == TRACE_ITER_RECORD_CMD) ||
5182 (mask == TRACE_ITER_TRACE_PRINTK))
5183 lockdep_assert_held(&event_mutex);
5184
5185 /* do nothing if flag is already set */
5186 if (!!(tr->trace_flags & mask) == !!enabled)
5187 return 0;
5188
5189 /* Give the tracer a chance to approve the change */
5190 if (tr->current_trace->flag_changed)
5191 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5192 return -EINVAL;
5193
5194 if (mask == TRACE_ITER_TRACE_PRINTK) {
5195 if (enabled) {
5196 update_printk_trace(tr);
5197 } else {
5198 /*
5199 * The global_trace cannot clear this.
5200 * It's flag only gets cleared if another instance sets it.
5201 */
5202 if (printk_trace == &global_trace)
5203 return -EINVAL;
5204 /*
5205 * An instance must always have it set.
5206 * by default, that's the global_trace instane.
5207 */
5208 if (printk_trace == tr)
5209 update_printk_trace(&global_trace);
5210 }
5211 }
5212
5213 if (enabled)
5214 tr->trace_flags |= mask;
5215 else
5216 tr->trace_flags &= ~mask;
5217
5218 if (mask == TRACE_ITER_RECORD_CMD)
5219 trace_event_enable_cmd_record(enabled);
5220
5221 if (mask == TRACE_ITER_RECORD_TGID) {
5222
5223 if (trace_alloc_tgid_map() < 0) {
5224 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5225 return -ENOMEM;
5226 }
5227
5228 trace_event_enable_tgid_record(enabled);
5229 }
5230
5231 if (mask == TRACE_ITER_EVENT_FORK)
5232 trace_event_follow_fork(tr, enabled);
5233
5234 if (mask == TRACE_ITER_FUNC_FORK)
5235 ftrace_pid_follow_fork(tr, enabled);
5236
5237 if (mask == TRACE_ITER_OVERWRITE) {
5238 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5239#ifdef CONFIG_TRACER_MAX_TRACE
5240 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5241#endif
5242 }
5243
5244 if (mask == TRACE_ITER_PRINTK) {
5245 trace_printk_start_stop_comm(enabled);
5246 trace_printk_control(enabled);
5247 }
5248
5249 return 0;
5250}
5251
5252int trace_set_options(struct trace_array *tr, char *option)
5253{
5254 char *cmp;
5255 int neg = 0;
5256 int ret;
5257 size_t orig_len = strlen(option);
5258 int len;
5259
5260 cmp = strstrip(option);
5261
5262 len = str_has_prefix(cmp, "no");
5263 if (len)
5264 neg = 1;
5265
5266 cmp += len;
5267
5268 mutex_lock(&event_mutex);
5269 mutex_lock(&trace_types_lock);
5270
5271 ret = match_string(trace_options, -1, cmp);
5272 /* If no option could be set, test the specific tracer options */
5273 if (ret < 0)
5274 ret = set_tracer_option(tr, cmp, neg);
5275 else
5276 ret = set_tracer_flag(tr, 1 << ret, !neg);
5277
5278 mutex_unlock(&trace_types_lock);
5279 mutex_unlock(&event_mutex);
5280
5281 /*
5282 * If the first trailing whitespace is replaced with '\0' by strstrip,
5283 * turn it back into a space.
5284 */
5285 if (orig_len > strlen(option))
5286 option[strlen(option)] = ' ';
5287
5288 return ret;
5289}
5290
5291static void __init apply_trace_boot_options(void)
5292{
5293 char *buf = trace_boot_options_buf;
5294 char *option;
5295
5296 while (true) {
5297 option = strsep(&buf, ",");
5298
5299 if (!option)
5300 break;
5301
5302 if (*option)
5303 trace_set_options(&global_trace, option);
5304
5305 /* Put back the comma to allow this to be called again */
5306 if (buf)
5307 *(buf - 1) = ',';
5308 }
5309}
5310
5311static ssize_t
5312tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5313 size_t cnt, loff_t *ppos)
5314{
5315 struct seq_file *m = filp->private_data;
5316 struct trace_array *tr = m->private;
5317 char buf[64];
5318 int ret;
5319
5320 if (cnt >= sizeof(buf))
5321 return -EINVAL;
5322
5323 if (copy_from_user(buf, ubuf, cnt))
5324 return -EFAULT;
5325
5326 buf[cnt] = 0;
5327
5328 ret = trace_set_options(tr, buf);
5329 if (ret < 0)
5330 return ret;
5331
5332 *ppos += cnt;
5333
5334 return cnt;
5335}
5336
5337static int tracing_trace_options_open(struct inode *inode, struct file *file)
5338{
5339 struct trace_array *tr = inode->i_private;
5340 int ret;
5341
5342 ret = tracing_check_open_get_tr(tr);
5343 if (ret)
5344 return ret;
5345
5346 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5347 if (ret < 0)
5348 trace_array_put(tr);
5349
5350 return ret;
5351}
5352
5353static const struct file_operations tracing_iter_fops = {
5354 .open = tracing_trace_options_open,
5355 .read = seq_read,
5356 .llseek = seq_lseek,
5357 .release = tracing_single_release_tr,
5358 .write = tracing_trace_options_write,
5359};
5360
5361static const char readme_msg[] =
5362 "tracing mini-HOWTO:\n\n"
5363 "By default tracefs removes all OTH file permission bits.\n"
5364 "When mounting tracefs an optional group id can be specified\n"
5365 "which adds the group to every directory and file in tracefs:\n\n"
5366 "\t e.g. mount -t tracefs [-o [gid=<gid>]] nodev /sys/kernel/tracing\n\n"
5367 "# echo 0 > tracing_on : quick way to disable tracing\n"
5368 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5369 " Important files:\n"
5370 " trace\t\t\t- The static contents of the buffer\n"
5371 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5372 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5373 " current_tracer\t- function and latency tracers\n"
5374 " available_tracers\t- list of configured tracers for current_tracer\n"
5375 " error_log\t- error log for failed commands (that support it)\n"
5376 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5377 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5378 " trace_clock\t\t- change the clock used to order events\n"
5379 " local: Per cpu clock but may not be synced across CPUs\n"
5380 " global: Synced across CPUs but slows tracing down.\n"
5381 " counter: Not a clock, but just an increment\n"
5382 " uptime: Jiffy counter from time of boot\n"
5383 " perf: Same clock that perf events use\n"
5384#ifdef CONFIG_X86_64
5385 " x86-tsc: TSC cycle counter\n"
5386#endif
5387 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5388 " delta: Delta difference against a buffer-wide timestamp\n"
5389 " absolute: Absolute (standalone) timestamp\n"
5390 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5391 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5392 " tracing_cpumask\t- Limit which CPUs to trace\n"
5393 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5394 "\t\t\t Remove sub-buffer with rmdir\n"
5395 " trace_options\t\t- Set format or modify how tracing happens\n"
5396 "\t\t\t Disable an option by prefixing 'no' to the\n"
5397 "\t\t\t option name\n"
5398 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5399#ifdef CONFIG_DYNAMIC_FTRACE
5400 "\n available_filter_functions - list of functions that can be filtered on\n"
5401 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5402 "\t\t\t functions\n"
5403 "\t accepts: func_full_name or glob-matching-pattern\n"
5404 "\t modules: Can select a group via module\n"
5405 "\t Format: :mod:<module-name>\n"
5406 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5407 "\t triggers: a command to perform when function is hit\n"
5408 "\t Format: <function>:<trigger>[:count]\n"
5409 "\t trigger: traceon, traceoff\n"
5410 "\t\t enable_event:<system>:<event>\n"
5411 "\t\t disable_event:<system>:<event>\n"
5412#ifdef CONFIG_STACKTRACE
5413 "\t\t stacktrace\n"
5414#endif
5415#ifdef CONFIG_TRACER_SNAPSHOT
5416 "\t\t snapshot\n"
5417#endif
5418 "\t\t dump\n"
5419 "\t\t cpudump\n"
5420 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5421 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5422 "\t The first one will disable tracing every time do_fault is hit\n"
5423 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5424 "\t The first time do trap is hit and it disables tracing, the\n"
5425 "\t counter will decrement to 2. If tracing is already disabled,\n"
5426 "\t the counter will not decrement. It only decrements when the\n"
5427 "\t trigger did work\n"
5428 "\t To remove trigger without count:\n"
5429 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5430 "\t To remove trigger with a count:\n"
5431 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5432 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5433 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5434 "\t modules: Can select a group via module command :mod:\n"
5435 "\t Does not accept triggers\n"
5436#endif /* CONFIG_DYNAMIC_FTRACE */
5437#ifdef CONFIG_FUNCTION_TRACER
5438 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5439 "\t\t (function)\n"
5440 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5441 "\t\t (function)\n"
5442#endif
5443#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5444 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5445 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5446 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5447#endif
5448#ifdef CONFIG_TRACER_SNAPSHOT
5449 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5450 "\t\t\t snapshot buffer. Read the contents for more\n"
5451 "\t\t\t information\n"
5452#endif
5453#ifdef CONFIG_STACK_TRACER
5454 " stack_trace\t\t- Shows the max stack trace when active\n"
5455 " stack_max_size\t- Shows current max stack size that was traced\n"
5456 "\t\t\t Write into this file to reset the max size (trigger a\n"
5457 "\t\t\t new trace)\n"
5458#ifdef CONFIG_DYNAMIC_FTRACE
5459 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5460 "\t\t\t traces\n"
5461#endif
5462#endif /* CONFIG_STACK_TRACER */
5463#ifdef CONFIG_DYNAMIC_EVENTS
5464 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5465 "\t\t\t Write into this file to define/undefine new trace events.\n"
5466#endif
5467#ifdef CONFIG_KPROBE_EVENTS
5468 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5469 "\t\t\t Write into this file to define/undefine new trace events.\n"
5470#endif
5471#ifdef CONFIG_UPROBE_EVENTS
5472 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5473 "\t\t\t Write into this file to define/undefine new trace events.\n"
5474#endif
5475#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5476 defined(CONFIG_FPROBE_EVENTS)
5477 "\t accepts: event-definitions (one definition per line)\n"
5478#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5479 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5480 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5481#endif
5482#ifdef CONFIG_FPROBE_EVENTS
5483 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5484 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5485#endif
5486#ifdef CONFIG_HIST_TRIGGERS
5487 "\t s:[synthetic/]<event> <field> [<field>]\n"
5488#endif
5489 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5490 "\t -:[<group>/][<event>]\n"
5491#ifdef CONFIG_KPROBE_EVENTS
5492 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5493 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5494#endif
5495#ifdef CONFIG_UPROBE_EVENTS
5496 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5497#endif
5498 "\t args: <name>=fetcharg[:type]\n"
5499 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5500#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5501 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5502#ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5503 "\t <argname>[->field[->field|.field...]],\n"
5504#endif
5505#else
5506 "\t $stack<index>, $stack, $retval, $comm,\n"
5507#endif
5508 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5509 "\t kernel return probes support: $retval, $arg<N>, $comm\n"
5510 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5511 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5512 "\t symstr, %pd/%pD, <type>\\[<array-size>\\]\n"
5513#ifdef CONFIG_HIST_TRIGGERS
5514 "\t field: <stype> <name>;\n"
5515 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5516 "\t [unsigned] char/int/long\n"
5517#endif
5518 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5519 "\t of the <attached-group>/<attached-event>.\n"
5520#endif
5521 " events/\t\t- Directory containing all trace event subsystems:\n"
5522 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5523 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5524 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5525 "\t\t\t events\n"
5526 " filter\t\t- If set, only events passing filter are traced\n"
5527 " events/<system>/<event>/\t- Directory containing control files for\n"
5528 "\t\t\t <event>:\n"
5529 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5530 " filter\t\t- If set, only events passing filter are traced\n"
5531 " trigger\t\t- If set, a command to perform when event is hit\n"
5532 "\t Format: <trigger>[:count][if <filter>]\n"
5533 "\t trigger: traceon, traceoff\n"
5534 "\t enable_event:<system>:<event>\n"
5535 "\t disable_event:<system>:<event>\n"
5536#ifdef CONFIG_HIST_TRIGGERS
5537 "\t enable_hist:<system>:<event>\n"
5538 "\t disable_hist:<system>:<event>\n"
5539#endif
5540#ifdef CONFIG_STACKTRACE
5541 "\t\t stacktrace\n"
5542#endif
5543#ifdef CONFIG_TRACER_SNAPSHOT
5544 "\t\t snapshot\n"
5545#endif
5546#ifdef CONFIG_HIST_TRIGGERS
5547 "\t\t hist (see below)\n"
5548#endif
5549 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5550 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5551 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5552 "\t events/block/block_unplug/trigger\n"
5553 "\t The first disables tracing every time block_unplug is hit.\n"
5554 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5555 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5556 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5557 "\t Like function triggers, the counter is only decremented if it\n"
5558 "\t enabled or disabled tracing.\n"
5559 "\t To remove a trigger without a count:\n"
5560 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5561 "\t To remove a trigger with a count:\n"
5562 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5563 "\t Filters can be ignored when removing a trigger.\n"
5564#ifdef CONFIG_HIST_TRIGGERS
5565 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5566 "\t Format: hist:keys=<field1[,field2,...]>\n"
5567 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5568 "\t [:values=<field1[,field2,...]>]\n"
5569 "\t [:sort=<field1[,field2,...]>]\n"
5570 "\t [:size=#entries]\n"
5571 "\t [:pause][:continue][:clear]\n"
5572 "\t [:name=histname1]\n"
5573 "\t [:nohitcount]\n"
5574 "\t [:<handler>.<action>]\n"
5575 "\t [if <filter>]\n\n"
5576 "\t Note, special fields can be used as well:\n"
5577 "\t common_timestamp - to record current timestamp\n"
5578 "\t common_cpu - to record the CPU the event happened on\n"
5579 "\n"
5580 "\t A hist trigger variable can be:\n"
5581 "\t - a reference to a field e.g. x=current_timestamp,\n"
5582 "\t - a reference to another variable e.g. y=$x,\n"
5583 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5584 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5585 "\n"
5586 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5587 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5588 "\t variable reference, field or numeric literal.\n"
5589 "\n"
5590 "\t When a matching event is hit, an entry is added to a hash\n"
5591 "\t table using the key(s) and value(s) named, and the value of a\n"
5592 "\t sum called 'hitcount' is incremented. Keys and values\n"
5593 "\t correspond to fields in the event's format description. Keys\n"
5594 "\t can be any field, or the special string 'common_stacktrace'.\n"
5595 "\t Compound keys consisting of up to two fields can be specified\n"
5596 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5597 "\t fields. Sort keys consisting of up to two fields can be\n"
5598 "\t specified using the 'sort' keyword. The sort direction can\n"
5599 "\t be modified by appending '.descending' or '.ascending' to a\n"
5600 "\t sort field. The 'size' parameter can be used to specify more\n"
5601 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5602 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5603 "\t its histogram data will be shared with other triggers of the\n"
5604 "\t same name, and trigger hits will update this common data.\n\n"
5605 "\t Reading the 'hist' file for the event will dump the hash\n"
5606 "\t table in its entirety to stdout. If there are multiple hist\n"
5607 "\t triggers attached to an event, there will be a table for each\n"
5608 "\t trigger in the output. The table displayed for a named\n"
5609 "\t trigger will be the same as any other instance having the\n"
5610 "\t same name. The default format used to display a given field\n"
5611 "\t can be modified by appending any of the following modifiers\n"
5612 "\t to the field name, as applicable:\n\n"
5613 "\t .hex display a number as a hex value\n"
5614 "\t .sym display an address as a symbol\n"
5615 "\t .sym-offset display an address as a symbol and offset\n"
5616 "\t .execname display a common_pid as a program name\n"
5617 "\t .syscall display a syscall id as a syscall name\n"
5618 "\t .log2 display log2 value rather than raw number\n"
5619 "\t .buckets=size display values in groups of size rather than raw number\n"
5620 "\t .usecs display a common_timestamp in microseconds\n"
5621 "\t .percent display a number of percentage value\n"
5622 "\t .graph display a bar-graph of a value\n\n"
5623 "\t The 'pause' parameter can be used to pause an existing hist\n"
5624 "\t trigger or to start a hist trigger but not log any events\n"
5625 "\t until told to do so. 'continue' can be used to start or\n"
5626 "\t restart a paused hist trigger.\n\n"
5627 "\t The 'clear' parameter will clear the contents of a running\n"
5628 "\t hist trigger and leave its current paused/active state\n"
5629 "\t unchanged.\n\n"
5630 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5631 "\t raw hitcount in the histogram.\n\n"
5632 "\t The enable_hist and disable_hist triggers can be used to\n"
5633 "\t have one event conditionally start and stop another event's\n"
5634 "\t already-attached hist trigger. The syntax is analogous to\n"
5635 "\t the enable_event and disable_event triggers.\n\n"
5636 "\t Hist trigger handlers and actions are executed whenever a\n"
5637 "\t a histogram entry is added or updated. They take the form:\n\n"
5638 "\t <handler>.<action>\n\n"
5639 "\t The available handlers are:\n\n"
5640 "\t onmatch(matching.event) - invoke on addition or update\n"
5641 "\t onmax(var) - invoke if var exceeds current max\n"
5642 "\t onchange(var) - invoke action if var changes\n\n"
5643 "\t The available actions are:\n\n"
5644 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5645 "\t save(field,...) - save current event fields\n"
5646#ifdef CONFIG_TRACER_SNAPSHOT
5647 "\t snapshot() - snapshot the trace buffer\n\n"
5648#endif
5649#ifdef CONFIG_SYNTH_EVENTS
5650 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5651 "\t Write into this file to define/undefine new synthetic events.\n"
5652 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5653#endif
5654#endif
5655;
5656
5657static ssize_t
5658tracing_readme_read(struct file *filp, char __user *ubuf,
5659 size_t cnt, loff_t *ppos)
5660{
5661 return simple_read_from_buffer(ubuf, cnt, ppos,
5662 readme_msg, strlen(readme_msg));
5663}
5664
5665static const struct file_operations tracing_readme_fops = {
5666 .open = tracing_open_generic,
5667 .read = tracing_readme_read,
5668 .llseek = generic_file_llseek,
5669};
5670
5671#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5672static union trace_eval_map_item *
5673update_eval_map(union trace_eval_map_item *ptr)
5674{
5675 if (!ptr->map.eval_string) {
5676 if (ptr->tail.next) {
5677 ptr = ptr->tail.next;
5678 /* Set ptr to the next real item (skip head) */
5679 ptr++;
5680 } else
5681 return NULL;
5682 }
5683 return ptr;
5684}
5685
5686static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5687{
5688 union trace_eval_map_item *ptr = v;
5689
5690 /*
5691 * Paranoid! If ptr points to end, we don't want to increment past it.
5692 * This really should never happen.
5693 */
5694 (*pos)++;
5695 ptr = update_eval_map(ptr);
5696 if (WARN_ON_ONCE(!ptr))
5697 return NULL;
5698
5699 ptr++;
5700 ptr = update_eval_map(ptr);
5701
5702 return ptr;
5703}
5704
5705static void *eval_map_start(struct seq_file *m, loff_t *pos)
5706{
5707 union trace_eval_map_item *v;
5708 loff_t l = 0;
5709
5710 mutex_lock(&trace_eval_mutex);
5711
5712 v = trace_eval_maps;
5713 if (v)
5714 v++;
5715
5716 while (v && l < *pos) {
5717 v = eval_map_next(m, v, &l);
5718 }
5719
5720 return v;
5721}
5722
5723static void eval_map_stop(struct seq_file *m, void *v)
5724{
5725 mutex_unlock(&trace_eval_mutex);
5726}
5727
5728static int eval_map_show(struct seq_file *m, void *v)
5729{
5730 union trace_eval_map_item *ptr = v;
5731
5732 seq_printf(m, "%s %ld (%s)\n",
5733 ptr->map.eval_string, ptr->map.eval_value,
5734 ptr->map.system);
5735
5736 return 0;
5737}
5738
5739static const struct seq_operations tracing_eval_map_seq_ops = {
5740 .start = eval_map_start,
5741 .next = eval_map_next,
5742 .stop = eval_map_stop,
5743 .show = eval_map_show,
5744};
5745
5746static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5747{
5748 int ret;
5749
5750 ret = tracing_check_open_get_tr(NULL);
5751 if (ret)
5752 return ret;
5753
5754 return seq_open(filp, &tracing_eval_map_seq_ops);
5755}
5756
5757static const struct file_operations tracing_eval_map_fops = {
5758 .open = tracing_eval_map_open,
5759 .read = seq_read,
5760 .llseek = seq_lseek,
5761 .release = seq_release,
5762};
5763
5764static inline union trace_eval_map_item *
5765trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5766{
5767 /* Return tail of array given the head */
5768 return ptr + ptr->head.length + 1;
5769}
5770
5771static void
5772trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5773 int len)
5774{
5775 struct trace_eval_map **stop;
5776 struct trace_eval_map **map;
5777 union trace_eval_map_item *map_array;
5778 union trace_eval_map_item *ptr;
5779
5780 stop = start + len;
5781
5782 /*
5783 * The trace_eval_maps contains the map plus a head and tail item,
5784 * where the head holds the module and length of array, and the
5785 * tail holds a pointer to the next list.
5786 */
5787 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5788 if (!map_array) {
5789 pr_warn("Unable to allocate trace eval mapping\n");
5790 return;
5791 }
5792
5793 guard(mutex)(&trace_eval_mutex);
5794
5795 if (!trace_eval_maps)
5796 trace_eval_maps = map_array;
5797 else {
5798 ptr = trace_eval_maps;
5799 for (;;) {
5800 ptr = trace_eval_jmp_to_tail(ptr);
5801 if (!ptr->tail.next)
5802 break;
5803 ptr = ptr->tail.next;
5804
5805 }
5806 ptr->tail.next = map_array;
5807 }
5808 map_array->head.mod = mod;
5809 map_array->head.length = len;
5810 map_array++;
5811
5812 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5813 map_array->map = **map;
5814 map_array++;
5815 }
5816 memset(map_array, 0, sizeof(*map_array));
5817}
5818
5819static void trace_create_eval_file(struct dentry *d_tracer)
5820{
5821 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
5822 NULL, &tracing_eval_map_fops);
5823}
5824
5825#else /* CONFIG_TRACE_EVAL_MAP_FILE */
5826static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5827static inline void trace_insert_eval_map_file(struct module *mod,
5828 struct trace_eval_map **start, int len) { }
5829#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5830
5831static void trace_insert_eval_map(struct module *mod,
5832 struct trace_eval_map **start, int len)
5833{
5834 struct trace_eval_map **map;
5835
5836 if (len <= 0)
5837 return;
5838
5839 map = start;
5840
5841 trace_event_eval_update(map, len);
5842
5843 trace_insert_eval_map_file(mod, start, len);
5844}
5845
5846static ssize_t
5847tracing_set_trace_read(struct file *filp, char __user *ubuf,
5848 size_t cnt, loff_t *ppos)
5849{
5850 struct trace_array *tr = filp->private_data;
5851 char buf[MAX_TRACER_SIZE+2];
5852 int r;
5853
5854 mutex_lock(&trace_types_lock);
5855 r = sprintf(buf, "%s\n", tr->current_trace->name);
5856 mutex_unlock(&trace_types_lock);
5857
5858 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5859}
5860
5861int tracer_init(struct tracer *t, struct trace_array *tr)
5862{
5863 tracing_reset_online_cpus(&tr->array_buffer);
5864 return t->init(tr);
5865}
5866
5867static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5868{
5869 int cpu;
5870
5871 for_each_tracing_cpu(cpu)
5872 per_cpu_ptr(buf->data, cpu)->entries = val;
5873}
5874
5875static void update_buffer_entries(struct array_buffer *buf, int cpu)
5876{
5877 if (cpu == RING_BUFFER_ALL_CPUS) {
5878 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
5879 } else {
5880 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
5881 }
5882}
5883
5884#ifdef CONFIG_TRACER_MAX_TRACE
5885/* resize @tr's buffer to the size of @size_tr's entries */
5886static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5887 struct array_buffer *size_buf, int cpu_id)
5888{
5889 int cpu, ret = 0;
5890
5891 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5892 for_each_tracing_cpu(cpu) {
5893 ret = ring_buffer_resize(trace_buf->buffer,
5894 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5895 if (ret < 0)
5896 break;
5897 per_cpu_ptr(trace_buf->data, cpu)->entries =
5898 per_cpu_ptr(size_buf->data, cpu)->entries;
5899 }
5900 } else {
5901 ret = ring_buffer_resize(trace_buf->buffer,
5902 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5903 if (ret == 0)
5904 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5905 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5906 }
5907
5908 return ret;
5909}
5910#endif /* CONFIG_TRACER_MAX_TRACE */
5911
5912static int __tracing_resize_ring_buffer(struct trace_array *tr,
5913 unsigned long size, int cpu)
5914{
5915 int ret;
5916
5917 /*
5918 * If kernel or user changes the size of the ring buffer
5919 * we use the size that was given, and we can forget about
5920 * expanding it later.
5921 */
5922 trace_set_ring_buffer_expanded(tr);
5923
5924 /* May be called before buffers are initialized */
5925 if (!tr->array_buffer.buffer)
5926 return 0;
5927
5928 /* Do not allow tracing while resizing ring buffer */
5929 tracing_stop_tr(tr);
5930
5931 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5932 if (ret < 0)
5933 goto out_start;
5934
5935#ifdef CONFIG_TRACER_MAX_TRACE
5936 if (!tr->allocated_snapshot)
5937 goto out;
5938
5939 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5940 if (ret < 0) {
5941 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5942 &tr->array_buffer, cpu);
5943 if (r < 0) {
5944 /*
5945 * AARGH! We are left with different
5946 * size max buffer!!!!
5947 * The max buffer is our "snapshot" buffer.
5948 * When a tracer needs a snapshot (one of the
5949 * latency tracers), it swaps the max buffer
5950 * with the saved snap shot. We succeeded to
5951 * update the size of the main buffer, but failed to
5952 * update the size of the max buffer. But when we tried
5953 * to reset the main buffer to the original size, we
5954 * failed there too. This is very unlikely to
5955 * happen, but if it does, warn and kill all
5956 * tracing.
5957 */
5958 WARN_ON(1);
5959 tracing_disabled = 1;
5960 }
5961 goto out_start;
5962 }
5963
5964 update_buffer_entries(&tr->max_buffer, cpu);
5965
5966 out:
5967#endif /* CONFIG_TRACER_MAX_TRACE */
5968
5969 update_buffer_entries(&tr->array_buffer, cpu);
5970 out_start:
5971 tracing_start_tr(tr);
5972 return ret;
5973}
5974
5975ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5976 unsigned long size, int cpu_id)
5977{
5978 guard(mutex)(&trace_types_lock);
5979
5980 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5981 /* make sure, this cpu is enabled in the mask */
5982 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask))
5983 return -EINVAL;
5984 }
5985
5986 return __tracing_resize_ring_buffer(tr, size, cpu_id);
5987}
5988
5989static void update_last_data(struct trace_array *tr)
5990{
5991 if (!tr->text_delta && !tr->data_delta)
5992 return;
5993
5994 /*
5995 * Need to clear all CPU buffers as there cannot be events
5996 * from the previous boot mixed with events with this boot
5997 * as that will cause a confusing trace. Need to clear all
5998 * CPU buffers, even for those that may currently be offline.
5999 */
6000 tracing_reset_all_cpus(&tr->array_buffer);
6001
6002 /* Using current data now */
6003 tr->text_delta = 0;
6004 tr->data_delta = 0;
6005}
6006
6007/**
6008 * tracing_update_buffers - used by tracing facility to expand ring buffers
6009 * @tr: The tracing instance
6010 *
6011 * To save on memory when the tracing is never used on a system with it
6012 * configured in. The ring buffers are set to a minimum size. But once
6013 * a user starts to use the tracing facility, then they need to grow
6014 * to their default size.
6015 *
6016 * This function is to be called when a tracer is about to be used.
6017 */
6018int tracing_update_buffers(struct trace_array *tr)
6019{
6020 int ret = 0;
6021
6022 mutex_lock(&trace_types_lock);
6023
6024 update_last_data(tr);
6025
6026 if (!tr->ring_buffer_expanded)
6027 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6028 RING_BUFFER_ALL_CPUS);
6029 mutex_unlock(&trace_types_lock);
6030
6031 return ret;
6032}
6033
6034struct trace_option_dentry;
6035
6036static void
6037create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6038
6039/*
6040 * Used to clear out the tracer before deletion of an instance.
6041 * Must have trace_types_lock held.
6042 */
6043static void tracing_set_nop(struct trace_array *tr)
6044{
6045 if (tr->current_trace == &nop_trace)
6046 return;
6047
6048 tr->current_trace->enabled--;
6049
6050 if (tr->current_trace->reset)
6051 tr->current_trace->reset(tr);
6052
6053 tr->current_trace = &nop_trace;
6054}
6055
6056static bool tracer_options_updated;
6057
6058static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6059{
6060 /* Only enable if the directory has been created already. */
6061 if (!tr->dir)
6062 return;
6063
6064 /* Only create trace option files after update_tracer_options finish */
6065 if (!tracer_options_updated)
6066 return;
6067
6068 create_trace_option_files(tr, t);
6069}
6070
6071int tracing_set_tracer(struct trace_array *tr, const char *buf)
6072{
6073 struct tracer *t;
6074#ifdef CONFIG_TRACER_MAX_TRACE
6075 bool had_max_tr;
6076#endif
6077 int ret;
6078
6079 guard(mutex)(&trace_types_lock);
6080
6081 update_last_data(tr);
6082
6083 if (!tr->ring_buffer_expanded) {
6084 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6085 RING_BUFFER_ALL_CPUS);
6086 if (ret < 0)
6087 return ret;
6088 ret = 0;
6089 }
6090
6091 for (t = trace_types; t; t = t->next) {
6092 if (strcmp(t->name, buf) == 0)
6093 break;
6094 }
6095 if (!t)
6096 return -EINVAL;
6097
6098 if (t == tr->current_trace)
6099 return 0;
6100
6101#ifdef CONFIG_TRACER_SNAPSHOT
6102 if (t->use_max_tr) {
6103 local_irq_disable();
6104 arch_spin_lock(&tr->max_lock);
6105 ret = tr->cond_snapshot ? -EBUSY : 0;
6106 arch_spin_unlock(&tr->max_lock);
6107 local_irq_enable();
6108 if (ret)
6109 return ret;
6110 }
6111#endif
6112 /* Some tracers won't work on kernel command line */
6113 if (system_state < SYSTEM_RUNNING && t->noboot) {
6114 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6115 t->name);
6116 return 0;
6117 }
6118
6119 /* Some tracers are only allowed for the top level buffer */
6120 if (!trace_ok_for_array(t, tr))
6121 return -EINVAL;
6122
6123 /* If trace pipe files are being read, we can't change the tracer */
6124 if (tr->trace_ref)
6125 return -EBUSY;
6126
6127 trace_branch_disable();
6128
6129 tr->current_trace->enabled--;
6130
6131 if (tr->current_trace->reset)
6132 tr->current_trace->reset(tr);
6133
6134#ifdef CONFIG_TRACER_MAX_TRACE
6135 had_max_tr = tr->current_trace->use_max_tr;
6136
6137 /* Current trace needs to be nop_trace before synchronize_rcu */
6138 tr->current_trace = &nop_trace;
6139
6140 if (had_max_tr && !t->use_max_tr) {
6141 /*
6142 * We need to make sure that the update_max_tr sees that
6143 * current_trace changed to nop_trace to keep it from
6144 * swapping the buffers after we resize it.
6145 * The update_max_tr is called from interrupts disabled
6146 * so a synchronized_sched() is sufficient.
6147 */
6148 synchronize_rcu();
6149 free_snapshot(tr);
6150 tracing_disarm_snapshot(tr);
6151 }
6152
6153 if (!had_max_tr && t->use_max_tr) {
6154 ret = tracing_arm_snapshot_locked(tr);
6155 if (ret)
6156 return ret;
6157 }
6158#else
6159 tr->current_trace = &nop_trace;
6160#endif
6161
6162 if (t->init) {
6163 ret = tracer_init(t, tr);
6164 if (ret) {
6165#ifdef CONFIG_TRACER_MAX_TRACE
6166 if (t->use_max_tr)
6167 tracing_disarm_snapshot(tr);
6168#endif
6169 return ret;
6170 }
6171 }
6172
6173 tr->current_trace = t;
6174 tr->current_trace->enabled++;
6175 trace_branch_enable(tr);
6176
6177 return 0;
6178}
6179
6180static ssize_t
6181tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6182 size_t cnt, loff_t *ppos)
6183{
6184 struct trace_array *tr = filp->private_data;
6185 char buf[MAX_TRACER_SIZE+1];
6186 char *name;
6187 size_t ret;
6188 int err;
6189
6190 ret = cnt;
6191
6192 if (cnt > MAX_TRACER_SIZE)
6193 cnt = MAX_TRACER_SIZE;
6194
6195 if (copy_from_user(buf, ubuf, cnt))
6196 return -EFAULT;
6197
6198 buf[cnt] = 0;
6199
6200 name = strim(buf);
6201
6202 err = tracing_set_tracer(tr, name);
6203 if (err)
6204 return err;
6205
6206 *ppos += ret;
6207
6208 return ret;
6209}
6210
6211static ssize_t
6212tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6213 size_t cnt, loff_t *ppos)
6214{
6215 char buf[64];
6216 int r;
6217
6218 r = snprintf(buf, sizeof(buf), "%ld\n",
6219 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6220 if (r > sizeof(buf))
6221 r = sizeof(buf);
6222 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6223}
6224
6225static ssize_t
6226tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6227 size_t cnt, loff_t *ppos)
6228{
6229 unsigned long val;
6230 int ret;
6231
6232 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6233 if (ret)
6234 return ret;
6235
6236 *ptr = val * 1000;
6237
6238 return cnt;
6239}
6240
6241static ssize_t
6242tracing_thresh_read(struct file *filp, char __user *ubuf,
6243 size_t cnt, loff_t *ppos)
6244{
6245 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6246}
6247
6248static ssize_t
6249tracing_thresh_write(struct file *filp, const char __user *ubuf,
6250 size_t cnt, loff_t *ppos)
6251{
6252 struct trace_array *tr = filp->private_data;
6253 int ret;
6254
6255 guard(mutex)(&trace_types_lock);
6256 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6257 if (ret < 0)
6258 return ret;
6259
6260 if (tr->current_trace->update_thresh) {
6261 ret = tr->current_trace->update_thresh(tr);
6262 if (ret < 0)
6263 return ret;
6264 }
6265
6266 return cnt;
6267}
6268
6269#ifdef CONFIG_TRACER_MAX_TRACE
6270
6271static ssize_t
6272tracing_max_lat_read(struct file *filp, char __user *ubuf,
6273 size_t cnt, loff_t *ppos)
6274{
6275 struct trace_array *tr = filp->private_data;
6276
6277 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6278}
6279
6280static ssize_t
6281tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6282 size_t cnt, loff_t *ppos)
6283{
6284 struct trace_array *tr = filp->private_data;
6285
6286 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6287}
6288
6289#endif
6290
6291static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6292{
6293 if (cpu == RING_BUFFER_ALL_CPUS) {
6294 if (cpumask_empty(tr->pipe_cpumask)) {
6295 cpumask_setall(tr->pipe_cpumask);
6296 return 0;
6297 }
6298 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6299 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6300 return 0;
6301 }
6302 return -EBUSY;
6303}
6304
6305static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6306{
6307 if (cpu == RING_BUFFER_ALL_CPUS) {
6308 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6309 cpumask_clear(tr->pipe_cpumask);
6310 } else {
6311 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6312 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6313 }
6314}
6315
6316static int tracing_open_pipe(struct inode *inode, struct file *filp)
6317{
6318 struct trace_array *tr = inode->i_private;
6319 struct trace_iterator *iter;
6320 int cpu;
6321 int ret;
6322
6323 ret = tracing_check_open_get_tr(tr);
6324 if (ret)
6325 return ret;
6326
6327 mutex_lock(&trace_types_lock);
6328 cpu = tracing_get_cpu(inode);
6329 ret = open_pipe_on_cpu(tr, cpu);
6330 if (ret)
6331 goto fail_pipe_on_cpu;
6332
6333 /* create a buffer to store the information to pass to userspace */
6334 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6335 if (!iter) {
6336 ret = -ENOMEM;
6337 goto fail_alloc_iter;
6338 }
6339
6340 trace_seq_init(&iter->seq);
6341 iter->trace = tr->current_trace;
6342
6343 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6344 ret = -ENOMEM;
6345 goto fail;
6346 }
6347
6348 /* trace pipe does not show start of buffer */
6349 cpumask_setall(iter->started);
6350
6351 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6352 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6353
6354 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6355 if (trace_clocks[tr->clock_id].in_ns)
6356 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6357
6358 iter->tr = tr;
6359 iter->array_buffer = &tr->array_buffer;
6360 iter->cpu_file = cpu;
6361 mutex_init(&iter->mutex);
6362 filp->private_data = iter;
6363
6364 if (iter->trace->pipe_open)
6365 iter->trace->pipe_open(iter);
6366
6367 nonseekable_open(inode, filp);
6368
6369 tr->trace_ref++;
6370
6371 mutex_unlock(&trace_types_lock);
6372 return ret;
6373
6374fail:
6375 kfree(iter);
6376fail_alloc_iter:
6377 close_pipe_on_cpu(tr, cpu);
6378fail_pipe_on_cpu:
6379 __trace_array_put(tr);
6380 mutex_unlock(&trace_types_lock);
6381 return ret;
6382}
6383
6384static int tracing_release_pipe(struct inode *inode, struct file *file)
6385{
6386 struct trace_iterator *iter = file->private_data;
6387 struct trace_array *tr = inode->i_private;
6388
6389 mutex_lock(&trace_types_lock);
6390
6391 tr->trace_ref--;
6392
6393 if (iter->trace->pipe_close)
6394 iter->trace->pipe_close(iter);
6395 close_pipe_on_cpu(tr, iter->cpu_file);
6396 mutex_unlock(&trace_types_lock);
6397
6398 free_trace_iter_content(iter);
6399 kfree(iter);
6400
6401 trace_array_put(tr);
6402
6403 return 0;
6404}
6405
6406static __poll_t
6407trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6408{
6409 struct trace_array *tr = iter->tr;
6410
6411 /* Iterators are static, they should be filled or empty */
6412 if (trace_buffer_iter(iter, iter->cpu_file))
6413 return EPOLLIN | EPOLLRDNORM;
6414
6415 if (tr->trace_flags & TRACE_ITER_BLOCK)
6416 /*
6417 * Always select as readable when in blocking mode
6418 */
6419 return EPOLLIN | EPOLLRDNORM;
6420 else
6421 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6422 filp, poll_table, iter->tr->buffer_percent);
6423}
6424
6425static __poll_t
6426tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6427{
6428 struct trace_iterator *iter = filp->private_data;
6429
6430 return trace_poll(iter, filp, poll_table);
6431}
6432
6433/* Must be called with iter->mutex held. */
6434static int tracing_wait_pipe(struct file *filp)
6435{
6436 struct trace_iterator *iter = filp->private_data;
6437 int ret;
6438
6439 while (trace_empty(iter)) {
6440
6441 if ((filp->f_flags & O_NONBLOCK)) {
6442 return -EAGAIN;
6443 }
6444
6445 /*
6446 * We block until we read something and tracing is disabled.
6447 * We still block if tracing is disabled, but we have never
6448 * read anything. This allows a user to cat this file, and
6449 * then enable tracing. But after we have read something,
6450 * we give an EOF when tracing is again disabled.
6451 *
6452 * iter->pos will be 0 if we haven't read anything.
6453 */
6454 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6455 break;
6456
6457 mutex_unlock(&iter->mutex);
6458
6459 ret = wait_on_pipe(iter, 0);
6460
6461 mutex_lock(&iter->mutex);
6462
6463 if (ret)
6464 return ret;
6465 }
6466
6467 return 1;
6468}
6469
6470/*
6471 * Consumer reader.
6472 */
6473static ssize_t
6474tracing_read_pipe(struct file *filp, char __user *ubuf,
6475 size_t cnt, loff_t *ppos)
6476{
6477 struct trace_iterator *iter = filp->private_data;
6478 ssize_t sret;
6479
6480 /*
6481 * Avoid more than one consumer on a single file descriptor
6482 * This is just a matter of traces coherency, the ring buffer itself
6483 * is protected.
6484 */
6485 guard(mutex)(&iter->mutex);
6486
6487 /* return any leftover data */
6488 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6489 if (sret != -EBUSY)
6490 return sret;
6491
6492 trace_seq_init(&iter->seq);
6493
6494 if (iter->trace->read) {
6495 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6496 if (sret)
6497 return sret;
6498 }
6499
6500waitagain:
6501 sret = tracing_wait_pipe(filp);
6502 if (sret <= 0)
6503 return sret;
6504
6505 /* stop when tracing is finished */
6506 if (trace_empty(iter))
6507 return 0;
6508
6509 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6510 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6511
6512 /* reset all but tr, trace, and overruns */
6513 trace_iterator_reset(iter);
6514 cpumask_clear(iter->started);
6515 trace_seq_init(&iter->seq);
6516
6517 trace_event_read_lock();
6518 trace_access_lock(iter->cpu_file);
6519 while (trace_find_next_entry_inc(iter) != NULL) {
6520 enum print_line_t ret;
6521 int save_len = iter->seq.seq.len;
6522
6523 ret = print_trace_line(iter);
6524 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6525 /*
6526 * If one print_trace_line() fills entire trace_seq in one shot,
6527 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6528 * In this case, we need to consume it, otherwise, loop will peek
6529 * this event next time, resulting in an infinite loop.
6530 */
6531 if (save_len == 0) {
6532 iter->seq.full = 0;
6533 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6534 trace_consume(iter);
6535 break;
6536 }
6537
6538 /* In other cases, don't print partial lines */
6539 iter->seq.seq.len = save_len;
6540 break;
6541 }
6542 if (ret != TRACE_TYPE_NO_CONSUME)
6543 trace_consume(iter);
6544
6545 if (trace_seq_used(&iter->seq) >= cnt)
6546 break;
6547
6548 /*
6549 * Setting the full flag means we reached the trace_seq buffer
6550 * size and we should leave by partial output condition above.
6551 * One of the trace_seq_* functions is not used properly.
6552 */
6553 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6554 iter->ent->type);
6555 }
6556 trace_access_unlock(iter->cpu_file);
6557 trace_event_read_unlock();
6558
6559 /* Now copy what we have to the user */
6560 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6561 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
6562 trace_seq_init(&iter->seq);
6563
6564 /*
6565 * If there was nothing to send to user, in spite of consuming trace
6566 * entries, go back to wait for more entries.
6567 */
6568 if (sret == -EBUSY)
6569 goto waitagain;
6570
6571 return sret;
6572}
6573
6574static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6575 unsigned int idx)
6576{
6577 __free_page(spd->pages[idx]);
6578}
6579
6580static size_t
6581tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6582{
6583 size_t count;
6584 int save_len;
6585 int ret;
6586
6587 /* Seq buffer is page-sized, exactly what we need. */
6588 for (;;) {
6589 save_len = iter->seq.seq.len;
6590 ret = print_trace_line(iter);
6591
6592 if (trace_seq_has_overflowed(&iter->seq)) {
6593 iter->seq.seq.len = save_len;
6594 break;
6595 }
6596
6597 /*
6598 * This should not be hit, because it should only
6599 * be set if the iter->seq overflowed. But check it
6600 * anyway to be safe.
6601 */
6602 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6603 iter->seq.seq.len = save_len;
6604 break;
6605 }
6606
6607 count = trace_seq_used(&iter->seq) - save_len;
6608 if (rem < count) {
6609 rem = 0;
6610 iter->seq.seq.len = save_len;
6611 break;
6612 }
6613
6614 if (ret != TRACE_TYPE_NO_CONSUME)
6615 trace_consume(iter);
6616 rem -= count;
6617 if (!trace_find_next_entry_inc(iter)) {
6618 rem = 0;
6619 iter->ent = NULL;
6620 break;
6621 }
6622 }
6623
6624 return rem;
6625}
6626
6627static ssize_t tracing_splice_read_pipe(struct file *filp,
6628 loff_t *ppos,
6629 struct pipe_inode_info *pipe,
6630 size_t len,
6631 unsigned int flags)
6632{
6633 struct page *pages_def[PIPE_DEF_BUFFERS];
6634 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6635 struct trace_iterator *iter = filp->private_data;
6636 struct splice_pipe_desc spd = {
6637 .pages = pages_def,
6638 .partial = partial_def,
6639 .nr_pages = 0, /* This gets updated below. */
6640 .nr_pages_max = PIPE_DEF_BUFFERS,
6641 .ops = &default_pipe_buf_ops,
6642 .spd_release = tracing_spd_release_pipe,
6643 };
6644 ssize_t ret;
6645 size_t rem;
6646 unsigned int i;
6647
6648 if (splice_grow_spd(pipe, &spd))
6649 return -ENOMEM;
6650
6651 mutex_lock(&iter->mutex);
6652
6653 if (iter->trace->splice_read) {
6654 ret = iter->trace->splice_read(iter, filp,
6655 ppos, pipe, len, flags);
6656 if (ret)
6657 goto out_err;
6658 }
6659
6660 ret = tracing_wait_pipe(filp);
6661 if (ret <= 0)
6662 goto out_err;
6663
6664 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6665 ret = -EFAULT;
6666 goto out_err;
6667 }
6668
6669 trace_event_read_lock();
6670 trace_access_lock(iter->cpu_file);
6671
6672 /* Fill as many pages as possible. */
6673 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6674 spd.pages[i] = alloc_page(GFP_KERNEL);
6675 if (!spd.pages[i])
6676 break;
6677
6678 rem = tracing_fill_pipe_page(rem, iter);
6679
6680 /* Copy the data into the page, so we can start over. */
6681 ret = trace_seq_to_buffer(&iter->seq,
6682 page_address(spd.pages[i]),
6683 trace_seq_used(&iter->seq));
6684 if (ret < 0) {
6685 __free_page(spd.pages[i]);
6686 break;
6687 }
6688 spd.partial[i].offset = 0;
6689 spd.partial[i].len = trace_seq_used(&iter->seq);
6690
6691 trace_seq_init(&iter->seq);
6692 }
6693
6694 trace_access_unlock(iter->cpu_file);
6695 trace_event_read_unlock();
6696 mutex_unlock(&iter->mutex);
6697
6698 spd.nr_pages = i;
6699
6700 if (i)
6701 ret = splice_to_pipe(pipe, &spd);
6702 else
6703 ret = 0;
6704out:
6705 splice_shrink_spd(&spd);
6706 return ret;
6707
6708out_err:
6709 mutex_unlock(&iter->mutex);
6710 goto out;
6711}
6712
6713static ssize_t
6714tracing_entries_read(struct file *filp, char __user *ubuf,
6715 size_t cnt, loff_t *ppos)
6716{
6717 struct inode *inode = file_inode(filp);
6718 struct trace_array *tr = inode->i_private;
6719 int cpu = tracing_get_cpu(inode);
6720 char buf[64];
6721 int r = 0;
6722 ssize_t ret;
6723
6724 mutex_lock(&trace_types_lock);
6725
6726 if (cpu == RING_BUFFER_ALL_CPUS) {
6727 int cpu, buf_size_same;
6728 unsigned long size;
6729
6730 size = 0;
6731 buf_size_same = 1;
6732 /* check if all cpu sizes are same */
6733 for_each_tracing_cpu(cpu) {
6734 /* fill in the size from first enabled cpu */
6735 if (size == 0)
6736 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6737 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6738 buf_size_same = 0;
6739 break;
6740 }
6741 }
6742
6743 if (buf_size_same) {
6744 if (!tr->ring_buffer_expanded)
6745 r = sprintf(buf, "%lu (expanded: %lu)\n",
6746 size >> 10,
6747 trace_buf_size >> 10);
6748 else
6749 r = sprintf(buf, "%lu\n", size >> 10);
6750 } else
6751 r = sprintf(buf, "X\n");
6752 } else
6753 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6754
6755 mutex_unlock(&trace_types_lock);
6756
6757 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6758 return ret;
6759}
6760
6761static ssize_t
6762tracing_entries_write(struct file *filp, const char __user *ubuf,
6763 size_t cnt, loff_t *ppos)
6764{
6765 struct inode *inode = file_inode(filp);
6766 struct trace_array *tr = inode->i_private;
6767 unsigned long val;
6768 int ret;
6769
6770 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6771 if (ret)
6772 return ret;
6773
6774 /* must have at least 1 entry */
6775 if (!val)
6776 return -EINVAL;
6777
6778 /* value is in KB */
6779 val <<= 10;
6780 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6781 if (ret < 0)
6782 return ret;
6783
6784 *ppos += cnt;
6785
6786 return cnt;
6787}
6788
6789static ssize_t
6790tracing_total_entries_read(struct file *filp, char __user *ubuf,
6791 size_t cnt, loff_t *ppos)
6792{
6793 struct trace_array *tr = filp->private_data;
6794 char buf[64];
6795 int r, cpu;
6796 unsigned long size = 0, expanded_size = 0;
6797
6798 mutex_lock(&trace_types_lock);
6799 for_each_tracing_cpu(cpu) {
6800 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6801 if (!tr->ring_buffer_expanded)
6802 expanded_size += trace_buf_size >> 10;
6803 }
6804 if (tr->ring_buffer_expanded)
6805 r = sprintf(buf, "%lu\n", size);
6806 else
6807 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6808 mutex_unlock(&trace_types_lock);
6809
6810 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6811}
6812
6813static ssize_t
6814tracing_last_boot_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
6815{
6816 struct trace_array *tr = filp->private_data;
6817 struct seq_buf seq;
6818 char buf[64];
6819
6820 seq_buf_init(&seq, buf, 64);
6821
6822 seq_buf_printf(&seq, "text delta:\t%ld\n", tr->text_delta);
6823 seq_buf_printf(&seq, "data delta:\t%ld\n", tr->data_delta);
6824
6825 return simple_read_from_buffer(ubuf, cnt, ppos, buf, seq_buf_used(&seq));
6826}
6827
6828static int tracing_buffer_meta_open(struct inode *inode, struct file *filp)
6829{
6830 struct trace_array *tr = inode->i_private;
6831 int cpu = tracing_get_cpu(inode);
6832 int ret;
6833
6834 ret = tracing_check_open_get_tr(tr);
6835 if (ret)
6836 return ret;
6837
6838 ret = ring_buffer_meta_seq_init(filp, tr->array_buffer.buffer, cpu);
6839 if (ret < 0)
6840 __trace_array_put(tr);
6841 return ret;
6842}
6843
6844static ssize_t
6845tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6846 size_t cnt, loff_t *ppos)
6847{
6848 /*
6849 * There is no need to read what the user has written, this function
6850 * is just to make sure that there is no error when "echo" is used
6851 */
6852
6853 *ppos += cnt;
6854
6855 return cnt;
6856}
6857
6858static int
6859tracing_free_buffer_release(struct inode *inode, struct file *filp)
6860{
6861 struct trace_array *tr = inode->i_private;
6862
6863 /* disable tracing ? */
6864 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6865 tracer_tracing_off(tr);
6866 /* resize the ring buffer to 0 */
6867 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6868
6869 trace_array_put(tr);
6870
6871 return 0;
6872}
6873
6874#define TRACE_MARKER_MAX_SIZE 4096
6875
6876static ssize_t
6877tracing_mark_write(struct file *filp, const char __user *ubuf,
6878 size_t cnt, loff_t *fpos)
6879{
6880 struct trace_array *tr = filp->private_data;
6881 struct ring_buffer_event *event;
6882 enum event_trigger_type tt = ETT_NONE;
6883 struct trace_buffer *buffer;
6884 struct print_entry *entry;
6885 int meta_size;
6886 ssize_t written;
6887 size_t size;
6888 int len;
6889
6890/* Used in tracing_mark_raw_write() as well */
6891#define FAULTED_STR "<faulted>"
6892#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6893
6894 if (tracing_disabled)
6895 return -EINVAL;
6896
6897 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6898 return -EINVAL;
6899
6900 if ((ssize_t)cnt < 0)
6901 return -EINVAL;
6902
6903 if (cnt > TRACE_MARKER_MAX_SIZE)
6904 cnt = TRACE_MARKER_MAX_SIZE;
6905
6906 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
6907 again:
6908 size = cnt + meta_size;
6909
6910 /* If less than "<faulted>", then make sure we can still add that */
6911 if (cnt < FAULTED_SIZE)
6912 size += FAULTED_SIZE - cnt;
6913
6914 buffer = tr->array_buffer.buffer;
6915 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6916 tracing_gen_ctx());
6917 if (unlikely(!event)) {
6918 /*
6919 * If the size was greater than what was allowed, then
6920 * make it smaller and try again.
6921 */
6922 if (size > ring_buffer_max_event_size(buffer)) {
6923 /* cnt < FAULTED size should never be bigger than max */
6924 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
6925 return -EBADF;
6926 cnt = ring_buffer_max_event_size(buffer) - meta_size;
6927 /* The above should only happen once */
6928 if (WARN_ON_ONCE(cnt + meta_size == size))
6929 return -EBADF;
6930 goto again;
6931 }
6932
6933 /* Ring buffer disabled, return as if not open for write */
6934 return -EBADF;
6935 }
6936
6937 entry = ring_buffer_event_data(event);
6938 entry->ip = _THIS_IP_;
6939
6940 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6941 if (len) {
6942 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6943 cnt = FAULTED_SIZE;
6944 written = -EFAULT;
6945 } else
6946 written = cnt;
6947
6948 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6949 /* do not add \n before testing triggers, but add \0 */
6950 entry->buf[cnt] = '\0';
6951 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
6952 }
6953
6954 if (entry->buf[cnt - 1] != '\n') {
6955 entry->buf[cnt] = '\n';
6956 entry->buf[cnt + 1] = '\0';
6957 } else
6958 entry->buf[cnt] = '\0';
6959
6960 if (static_branch_unlikely(&trace_marker_exports_enabled))
6961 ftrace_exports(event, TRACE_EXPORT_MARKER);
6962 __buffer_unlock_commit(buffer, event);
6963
6964 if (tt)
6965 event_triggers_post_call(tr->trace_marker_file, tt);
6966
6967 return written;
6968}
6969
6970static ssize_t
6971tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6972 size_t cnt, loff_t *fpos)
6973{
6974 struct trace_array *tr = filp->private_data;
6975 struct ring_buffer_event *event;
6976 struct trace_buffer *buffer;
6977 struct raw_data_entry *entry;
6978 ssize_t written;
6979 int size;
6980 int len;
6981
6982#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6983
6984 if (tracing_disabled)
6985 return -EINVAL;
6986
6987 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6988 return -EINVAL;
6989
6990 /* The marker must at least have a tag id */
6991 if (cnt < sizeof(unsigned int))
6992 return -EINVAL;
6993
6994 size = sizeof(*entry) + cnt;
6995 if (cnt < FAULT_SIZE_ID)
6996 size += FAULT_SIZE_ID - cnt;
6997
6998 buffer = tr->array_buffer.buffer;
6999
7000 if (size > ring_buffer_max_event_size(buffer))
7001 return -EINVAL;
7002
7003 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7004 tracing_gen_ctx());
7005 if (!event)
7006 /* Ring buffer disabled, return as if not open for write */
7007 return -EBADF;
7008
7009 entry = ring_buffer_event_data(event);
7010
7011 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7012 if (len) {
7013 entry->id = -1;
7014 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7015 written = -EFAULT;
7016 } else
7017 written = cnt;
7018
7019 __buffer_unlock_commit(buffer, event);
7020
7021 return written;
7022}
7023
7024static int tracing_clock_show(struct seq_file *m, void *v)
7025{
7026 struct trace_array *tr = m->private;
7027 int i;
7028
7029 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7030 seq_printf(m,
7031 "%s%s%s%s", i ? " " : "",
7032 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7033 i == tr->clock_id ? "]" : "");
7034 seq_putc(m, '\n');
7035
7036 return 0;
7037}
7038
7039int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7040{
7041 int i;
7042
7043 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7044 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7045 break;
7046 }
7047 if (i == ARRAY_SIZE(trace_clocks))
7048 return -EINVAL;
7049
7050 mutex_lock(&trace_types_lock);
7051
7052 tr->clock_id = i;
7053
7054 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7055
7056 /*
7057 * New clock may not be consistent with the previous clock.
7058 * Reset the buffer so that it doesn't have incomparable timestamps.
7059 */
7060 tracing_reset_online_cpus(&tr->array_buffer);
7061
7062#ifdef CONFIG_TRACER_MAX_TRACE
7063 if (tr->max_buffer.buffer)
7064 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7065 tracing_reset_online_cpus(&tr->max_buffer);
7066#endif
7067
7068 mutex_unlock(&trace_types_lock);
7069
7070 return 0;
7071}
7072
7073static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7074 size_t cnt, loff_t *fpos)
7075{
7076 struct seq_file *m = filp->private_data;
7077 struct trace_array *tr = m->private;
7078 char buf[64];
7079 const char *clockstr;
7080 int ret;
7081
7082 if (cnt >= sizeof(buf))
7083 return -EINVAL;
7084
7085 if (copy_from_user(buf, ubuf, cnt))
7086 return -EFAULT;
7087
7088 buf[cnt] = 0;
7089
7090 clockstr = strstrip(buf);
7091
7092 ret = tracing_set_clock(tr, clockstr);
7093 if (ret)
7094 return ret;
7095
7096 *fpos += cnt;
7097
7098 return cnt;
7099}
7100
7101static int tracing_clock_open(struct inode *inode, struct file *file)
7102{
7103 struct trace_array *tr = inode->i_private;
7104 int ret;
7105
7106 ret = tracing_check_open_get_tr(tr);
7107 if (ret)
7108 return ret;
7109
7110 ret = single_open(file, tracing_clock_show, inode->i_private);
7111 if (ret < 0)
7112 trace_array_put(tr);
7113
7114 return ret;
7115}
7116
7117static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7118{
7119 struct trace_array *tr = m->private;
7120
7121 mutex_lock(&trace_types_lock);
7122
7123 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7124 seq_puts(m, "delta [absolute]\n");
7125 else
7126 seq_puts(m, "[delta] absolute\n");
7127
7128 mutex_unlock(&trace_types_lock);
7129
7130 return 0;
7131}
7132
7133static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7134{
7135 struct trace_array *tr = inode->i_private;
7136 int ret;
7137
7138 ret = tracing_check_open_get_tr(tr);
7139 if (ret)
7140 return ret;
7141
7142 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7143 if (ret < 0)
7144 trace_array_put(tr);
7145
7146 return ret;
7147}
7148
7149u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7150{
7151 if (rbe == this_cpu_read(trace_buffered_event))
7152 return ring_buffer_time_stamp(buffer);
7153
7154 return ring_buffer_event_time_stamp(buffer, rbe);
7155}
7156
7157/*
7158 * Set or disable using the per CPU trace_buffer_event when possible.
7159 */
7160int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7161{
7162 guard(mutex)(&trace_types_lock);
7163
7164 if (set && tr->no_filter_buffering_ref++)
7165 return 0;
7166
7167 if (!set) {
7168 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref))
7169 return -EINVAL;
7170
7171 --tr->no_filter_buffering_ref;
7172 }
7173
7174 return 0;
7175}
7176
7177struct ftrace_buffer_info {
7178 struct trace_iterator iter;
7179 void *spare;
7180 unsigned int spare_cpu;
7181 unsigned int spare_size;
7182 unsigned int read;
7183};
7184
7185#ifdef CONFIG_TRACER_SNAPSHOT
7186static int tracing_snapshot_open(struct inode *inode, struct file *file)
7187{
7188 struct trace_array *tr = inode->i_private;
7189 struct trace_iterator *iter;
7190 struct seq_file *m;
7191 int ret;
7192
7193 ret = tracing_check_open_get_tr(tr);
7194 if (ret)
7195 return ret;
7196
7197 if (file->f_mode & FMODE_READ) {
7198 iter = __tracing_open(inode, file, true);
7199 if (IS_ERR(iter))
7200 ret = PTR_ERR(iter);
7201 } else {
7202 /* Writes still need the seq_file to hold the private data */
7203 ret = -ENOMEM;
7204 m = kzalloc(sizeof(*m), GFP_KERNEL);
7205 if (!m)
7206 goto out;
7207 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7208 if (!iter) {
7209 kfree(m);
7210 goto out;
7211 }
7212 ret = 0;
7213
7214 iter->tr = tr;
7215 iter->array_buffer = &tr->max_buffer;
7216 iter->cpu_file = tracing_get_cpu(inode);
7217 m->private = iter;
7218 file->private_data = m;
7219 }
7220out:
7221 if (ret < 0)
7222 trace_array_put(tr);
7223
7224 return ret;
7225}
7226
7227static void tracing_swap_cpu_buffer(void *tr)
7228{
7229 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7230}
7231
7232static ssize_t
7233tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7234 loff_t *ppos)
7235{
7236 struct seq_file *m = filp->private_data;
7237 struct trace_iterator *iter = m->private;
7238 struct trace_array *tr = iter->tr;
7239 unsigned long val;
7240 int ret;
7241
7242 ret = tracing_update_buffers(tr);
7243 if (ret < 0)
7244 return ret;
7245
7246 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7247 if (ret)
7248 return ret;
7249
7250 guard(mutex)(&trace_types_lock);
7251
7252 if (tr->current_trace->use_max_tr)
7253 return -EBUSY;
7254
7255 local_irq_disable();
7256 arch_spin_lock(&tr->max_lock);
7257 if (tr->cond_snapshot)
7258 ret = -EBUSY;
7259 arch_spin_unlock(&tr->max_lock);
7260 local_irq_enable();
7261 if (ret)
7262 return ret;
7263
7264 switch (val) {
7265 case 0:
7266 if (iter->cpu_file != RING_BUFFER_ALL_CPUS)
7267 return -EINVAL;
7268 if (tr->allocated_snapshot)
7269 free_snapshot(tr);
7270 break;
7271 case 1:
7272/* Only allow per-cpu swap if the ring buffer supports it */
7273#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7274 if (iter->cpu_file != RING_BUFFER_ALL_CPUS)
7275 return -EINVAL;
7276#endif
7277 if (tr->allocated_snapshot)
7278 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7279 &tr->array_buffer, iter->cpu_file);
7280
7281 ret = tracing_arm_snapshot_locked(tr);
7282 if (ret)
7283 return ret;
7284
7285 /* Now, we're going to swap */
7286 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7287 local_irq_disable();
7288 update_max_tr(tr, current, smp_processor_id(), NULL);
7289 local_irq_enable();
7290 } else {
7291 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7292 (void *)tr, 1);
7293 }
7294 tracing_disarm_snapshot(tr);
7295 break;
7296 default:
7297 if (tr->allocated_snapshot) {
7298 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7299 tracing_reset_online_cpus(&tr->max_buffer);
7300 else
7301 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7302 }
7303 break;
7304 }
7305
7306 if (ret >= 0) {
7307 *ppos += cnt;
7308 ret = cnt;
7309 }
7310
7311 return ret;
7312}
7313
7314static int tracing_snapshot_release(struct inode *inode, struct file *file)
7315{
7316 struct seq_file *m = file->private_data;
7317 int ret;
7318
7319 ret = tracing_release(inode, file);
7320
7321 if (file->f_mode & FMODE_READ)
7322 return ret;
7323
7324 /* If write only, the seq_file is just a stub */
7325 if (m)
7326 kfree(m->private);
7327 kfree(m);
7328
7329 return 0;
7330}
7331
7332static int tracing_buffers_open(struct inode *inode, struct file *filp);
7333static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7334 size_t count, loff_t *ppos);
7335static int tracing_buffers_release(struct inode *inode, struct file *file);
7336static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7337 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7338
7339static int snapshot_raw_open(struct inode *inode, struct file *filp)
7340{
7341 struct ftrace_buffer_info *info;
7342 int ret;
7343
7344 /* The following checks for tracefs lockdown */
7345 ret = tracing_buffers_open(inode, filp);
7346 if (ret < 0)
7347 return ret;
7348
7349 info = filp->private_data;
7350
7351 if (info->iter.trace->use_max_tr) {
7352 tracing_buffers_release(inode, filp);
7353 return -EBUSY;
7354 }
7355
7356 info->iter.snapshot = true;
7357 info->iter.array_buffer = &info->iter.tr->max_buffer;
7358
7359 return ret;
7360}
7361
7362#endif /* CONFIG_TRACER_SNAPSHOT */
7363
7364
7365static const struct file_operations tracing_thresh_fops = {
7366 .open = tracing_open_generic,
7367 .read = tracing_thresh_read,
7368 .write = tracing_thresh_write,
7369 .llseek = generic_file_llseek,
7370};
7371
7372#ifdef CONFIG_TRACER_MAX_TRACE
7373static const struct file_operations tracing_max_lat_fops = {
7374 .open = tracing_open_generic_tr,
7375 .read = tracing_max_lat_read,
7376 .write = tracing_max_lat_write,
7377 .llseek = generic_file_llseek,
7378 .release = tracing_release_generic_tr,
7379};
7380#endif
7381
7382static const struct file_operations set_tracer_fops = {
7383 .open = tracing_open_generic_tr,
7384 .read = tracing_set_trace_read,
7385 .write = tracing_set_trace_write,
7386 .llseek = generic_file_llseek,
7387 .release = tracing_release_generic_tr,
7388};
7389
7390static const struct file_operations tracing_pipe_fops = {
7391 .open = tracing_open_pipe,
7392 .poll = tracing_poll_pipe,
7393 .read = tracing_read_pipe,
7394 .splice_read = tracing_splice_read_pipe,
7395 .release = tracing_release_pipe,
7396};
7397
7398static const struct file_operations tracing_entries_fops = {
7399 .open = tracing_open_generic_tr,
7400 .read = tracing_entries_read,
7401 .write = tracing_entries_write,
7402 .llseek = generic_file_llseek,
7403 .release = tracing_release_generic_tr,
7404};
7405
7406static const struct file_operations tracing_buffer_meta_fops = {
7407 .open = tracing_buffer_meta_open,
7408 .read = seq_read,
7409 .llseek = seq_lseek,
7410 .release = tracing_seq_release,
7411};
7412
7413static const struct file_operations tracing_total_entries_fops = {
7414 .open = tracing_open_generic_tr,
7415 .read = tracing_total_entries_read,
7416 .llseek = generic_file_llseek,
7417 .release = tracing_release_generic_tr,
7418};
7419
7420static const struct file_operations tracing_free_buffer_fops = {
7421 .open = tracing_open_generic_tr,
7422 .write = tracing_free_buffer_write,
7423 .release = tracing_free_buffer_release,
7424};
7425
7426static const struct file_operations tracing_mark_fops = {
7427 .open = tracing_mark_open,
7428 .write = tracing_mark_write,
7429 .release = tracing_release_generic_tr,
7430};
7431
7432static const struct file_operations tracing_mark_raw_fops = {
7433 .open = tracing_mark_open,
7434 .write = tracing_mark_raw_write,
7435 .release = tracing_release_generic_tr,
7436};
7437
7438static const struct file_operations trace_clock_fops = {
7439 .open = tracing_clock_open,
7440 .read = seq_read,
7441 .llseek = seq_lseek,
7442 .release = tracing_single_release_tr,
7443 .write = tracing_clock_write,
7444};
7445
7446static const struct file_operations trace_time_stamp_mode_fops = {
7447 .open = tracing_time_stamp_mode_open,
7448 .read = seq_read,
7449 .llseek = seq_lseek,
7450 .release = tracing_single_release_tr,
7451};
7452
7453static const struct file_operations last_boot_fops = {
7454 .open = tracing_open_generic_tr,
7455 .read = tracing_last_boot_read,
7456 .llseek = generic_file_llseek,
7457 .release = tracing_release_generic_tr,
7458};
7459
7460#ifdef CONFIG_TRACER_SNAPSHOT
7461static const struct file_operations snapshot_fops = {
7462 .open = tracing_snapshot_open,
7463 .read = seq_read,
7464 .write = tracing_snapshot_write,
7465 .llseek = tracing_lseek,
7466 .release = tracing_snapshot_release,
7467};
7468
7469static const struct file_operations snapshot_raw_fops = {
7470 .open = snapshot_raw_open,
7471 .read = tracing_buffers_read,
7472 .release = tracing_buffers_release,
7473 .splice_read = tracing_buffers_splice_read,
7474};
7475
7476#endif /* CONFIG_TRACER_SNAPSHOT */
7477
7478/*
7479 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7480 * @filp: The active open file structure
7481 * @ubuf: The userspace provided buffer to read value into
7482 * @cnt: The maximum number of bytes to read
7483 * @ppos: The current "file" position
7484 *
7485 * This function implements the write interface for a struct trace_min_max_param.
7486 * The filp->private_data must point to a trace_min_max_param structure that
7487 * defines where to write the value, the min and the max acceptable values,
7488 * and a lock to protect the write.
7489 */
7490static ssize_t
7491trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7492{
7493 struct trace_min_max_param *param = filp->private_data;
7494 u64 val;
7495 int err;
7496
7497 if (!param)
7498 return -EFAULT;
7499
7500 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7501 if (err)
7502 return err;
7503
7504 if (param->lock)
7505 mutex_lock(param->lock);
7506
7507 if (param->min && val < *param->min)
7508 err = -EINVAL;
7509
7510 if (param->max && val > *param->max)
7511 err = -EINVAL;
7512
7513 if (!err)
7514 *param->val = val;
7515
7516 if (param->lock)
7517 mutex_unlock(param->lock);
7518
7519 if (err)
7520 return err;
7521
7522 return cnt;
7523}
7524
7525/*
7526 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7527 * @filp: The active open file structure
7528 * @ubuf: The userspace provided buffer to read value into
7529 * @cnt: The maximum number of bytes to read
7530 * @ppos: The current "file" position
7531 *
7532 * This function implements the read interface for a struct trace_min_max_param.
7533 * The filp->private_data must point to a trace_min_max_param struct with valid
7534 * data.
7535 */
7536static ssize_t
7537trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7538{
7539 struct trace_min_max_param *param = filp->private_data;
7540 char buf[U64_STR_SIZE];
7541 int len;
7542 u64 val;
7543
7544 if (!param)
7545 return -EFAULT;
7546
7547 val = *param->val;
7548
7549 if (cnt > sizeof(buf))
7550 cnt = sizeof(buf);
7551
7552 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7553
7554 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7555}
7556
7557const struct file_operations trace_min_max_fops = {
7558 .open = tracing_open_generic,
7559 .read = trace_min_max_read,
7560 .write = trace_min_max_write,
7561};
7562
7563#define TRACING_LOG_ERRS_MAX 8
7564#define TRACING_LOG_LOC_MAX 128
7565
7566#define CMD_PREFIX " Command: "
7567
7568struct err_info {
7569 const char **errs; /* ptr to loc-specific array of err strings */
7570 u8 type; /* index into errs -> specific err string */
7571 u16 pos; /* caret position */
7572 u64 ts;
7573};
7574
7575struct tracing_log_err {
7576 struct list_head list;
7577 struct err_info info;
7578 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7579 char *cmd; /* what caused err */
7580};
7581
7582static DEFINE_MUTEX(tracing_err_log_lock);
7583
7584static struct tracing_log_err *alloc_tracing_log_err(int len)
7585{
7586 struct tracing_log_err *err;
7587
7588 err = kzalloc(sizeof(*err), GFP_KERNEL);
7589 if (!err)
7590 return ERR_PTR(-ENOMEM);
7591
7592 err->cmd = kzalloc(len, GFP_KERNEL);
7593 if (!err->cmd) {
7594 kfree(err);
7595 return ERR_PTR(-ENOMEM);
7596 }
7597
7598 return err;
7599}
7600
7601static void free_tracing_log_err(struct tracing_log_err *err)
7602{
7603 kfree(err->cmd);
7604 kfree(err);
7605}
7606
7607static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7608 int len)
7609{
7610 struct tracing_log_err *err;
7611 char *cmd;
7612
7613 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7614 err = alloc_tracing_log_err(len);
7615 if (PTR_ERR(err) != -ENOMEM)
7616 tr->n_err_log_entries++;
7617
7618 return err;
7619 }
7620 cmd = kzalloc(len, GFP_KERNEL);
7621 if (!cmd)
7622 return ERR_PTR(-ENOMEM);
7623 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7624 kfree(err->cmd);
7625 err->cmd = cmd;
7626 list_del(&err->list);
7627
7628 return err;
7629}
7630
7631/**
7632 * err_pos - find the position of a string within a command for error careting
7633 * @cmd: The tracing command that caused the error
7634 * @str: The string to position the caret at within @cmd
7635 *
7636 * Finds the position of the first occurrence of @str within @cmd. The
7637 * return value can be passed to tracing_log_err() for caret placement
7638 * within @cmd.
7639 *
7640 * Returns the index within @cmd of the first occurrence of @str or 0
7641 * if @str was not found.
7642 */
7643unsigned int err_pos(char *cmd, const char *str)
7644{
7645 char *found;
7646
7647 if (WARN_ON(!strlen(cmd)))
7648 return 0;
7649
7650 found = strstr(cmd, str);
7651 if (found)
7652 return found - cmd;
7653
7654 return 0;
7655}
7656
7657/**
7658 * tracing_log_err - write an error to the tracing error log
7659 * @tr: The associated trace array for the error (NULL for top level array)
7660 * @loc: A string describing where the error occurred
7661 * @cmd: The tracing command that caused the error
7662 * @errs: The array of loc-specific static error strings
7663 * @type: The index into errs[], which produces the specific static err string
7664 * @pos: The position the caret should be placed in the cmd
7665 *
7666 * Writes an error into tracing/error_log of the form:
7667 *
7668 * <loc>: error: <text>
7669 * Command: <cmd>
7670 * ^
7671 *
7672 * tracing/error_log is a small log file containing the last
7673 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7674 * unless there has been a tracing error, and the error log can be
7675 * cleared and have its memory freed by writing the empty string in
7676 * truncation mode to it i.e. echo > tracing/error_log.
7677 *
7678 * NOTE: the @errs array along with the @type param are used to
7679 * produce a static error string - this string is not copied and saved
7680 * when the error is logged - only a pointer to it is saved. See
7681 * existing callers for examples of how static strings are typically
7682 * defined for use with tracing_log_err().
7683 */
7684void tracing_log_err(struct trace_array *tr,
7685 const char *loc, const char *cmd,
7686 const char **errs, u8 type, u16 pos)
7687{
7688 struct tracing_log_err *err;
7689 int len = 0;
7690
7691 if (!tr)
7692 tr = &global_trace;
7693
7694 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7695
7696 guard(mutex)(&tracing_err_log_lock);
7697
7698 err = get_tracing_log_err(tr, len);
7699 if (PTR_ERR(err) == -ENOMEM)
7700 return;
7701
7702 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7703 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7704
7705 err->info.errs = errs;
7706 err->info.type = type;
7707 err->info.pos = pos;
7708 err->info.ts = local_clock();
7709
7710 list_add_tail(&err->list, &tr->err_log);
7711}
7712
7713static void clear_tracing_err_log(struct trace_array *tr)
7714{
7715 struct tracing_log_err *err, *next;
7716
7717 mutex_lock(&tracing_err_log_lock);
7718 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7719 list_del(&err->list);
7720 free_tracing_log_err(err);
7721 }
7722
7723 tr->n_err_log_entries = 0;
7724 mutex_unlock(&tracing_err_log_lock);
7725}
7726
7727static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7728{
7729 struct trace_array *tr = m->private;
7730
7731 mutex_lock(&tracing_err_log_lock);
7732
7733 return seq_list_start(&tr->err_log, *pos);
7734}
7735
7736static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7737{
7738 struct trace_array *tr = m->private;
7739
7740 return seq_list_next(v, &tr->err_log, pos);
7741}
7742
7743static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7744{
7745 mutex_unlock(&tracing_err_log_lock);
7746}
7747
7748static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7749{
7750 u16 i;
7751
7752 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7753 seq_putc(m, ' ');
7754 for (i = 0; i < pos; i++)
7755 seq_putc(m, ' ');
7756 seq_puts(m, "^\n");
7757}
7758
7759static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7760{
7761 struct tracing_log_err *err = v;
7762
7763 if (err) {
7764 const char *err_text = err->info.errs[err->info.type];
7765 u64 sec = err->info.ts;
7766 u32 nsec;
7767
7768 nsec = do_div(sec, NSEC_PER_SEC);
7769 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7770 err->loc, err_text);
7771 seq_printf(m, "%s", err->cmd);
7772 tracing_err_log_show_pos(m, err->info.pos);
7773 }
7774
7775 return 0;
7776}
7777
7778static const struct seq_operations tracing_err_log_seq_ops = {
7779 .start = tracing_err_log_seq_start,
7780 .next = tracing_err_log_seq_next,
7781 .stop = tracing_err_log_seq_stop,
7782 .show = tracing_err_log_seq_show
7783};
7784
7785static int tracing_err_log_open(struct inode *inode, struct file *file)
7786{
7787 struct trace_array *tr = inode->i_private;
7788 int ret = 0;
7789
7790 ret = tracing_check_open_get_tr(tr);
7791 if (ret)
7792 return ret;
7793
7794 /* If this file was opened for write, then erase contents */
7795 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7796 clear_tracing_err_log(tr);
7797
7798 if (file->f_mode & FMODE_READ) {
7799 ret = seq_open(file, &tracing_err_log_seq_ops);
7800 if (!ret) {
7801 struct seq_file *m = file->private_data;
7802 m->private = tr;
7803 } else {
7804 trace_array_put(tr);
7805 }
7806 }
7807 return ret;
7808}
7809
7810static ssize_t tracing_err_log_write(struct file *file,
7811 const char __user *buffer,
7812 size_t count, loff_t *ppos)
7813{
7814 return count;
7815}
7816
7817static int tracing_err_log_release(struct inode *inode, struct file *file)
7818{
7819 struct trace_array *tr = inode->i_private;
7820
7821 trace_array_put(tr);
7822
7823 if (file->f_mode & FMODE_READ)
7824 seq_release(inode, file);
7825
7826 return 0;
7827}
7828
7829static const struct file_operations tracing_err_log_fops = {
7830 .open = tracing_err_log_open,
7831 .write = tracing_err_log_write,
7832 .read = seq_read,
7833 .llseek = tracing_lseek,
7834 .release = tracing_err_log_release,
7835};
7836
7837static int tracing_buffers_open(struct inode *inode, struct file *filp)
7838{
7839 struct trace_array *tr = inode->i_private;
7840 struct ftrace_buffer_info *info;
7841 int ret;
7842
7843 ret = tracing_check_open_get_tr(tr);
7844 if (ret)
7845 return ret;
7846
7847 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7848 if (!info) {
7849 trace_array_put(tr);
7850 return -ENOMEM;
7851 }
7852
7853 mutex_lock(&trace_types_lock);
7854
7855 info->iter.tr = tr;
7856 info->iter.cpu_file = tracing_get_cpu(inode);
7857 info->iter.trace = tr->current_trace;
7858 info->iter.array_buffer = &tr->array_buffer;
7859 info->spare = NULL;
7860 /* Force reading ring buffer for first read */
7861 info->read = (unsigned int)-1;
7862
7863 filp->private_data = info;
7864
7865 tr->trace_ref++;
7866
7867 mutex_unlock(&trace_types_lock);
7868
7869 ret = nonseekable_open(inode, filp);
7870 if (ret < 0)
7871 trace_array_put(tr);
7872
7873 return ret;
7874}
7875
7876static __poll_t
7877tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7878{
7879 struct ftrace_buffer_info *info = filp->private_data;
7880 struct trace_iterator *iter = &info->iter;
7881
7882 return trace_poll(iter, filp, poll_table);
7883}
7884
7885static ssize_t
7886tracing_buffers_read(struct file *filp, char __user *ubuf,
7887 size_t count, loff_t *ppos)
7888{
7889 struct ftrace_buffer_info *info = filp->private_data;
7890 struct trace_iterator *iter = &info->iter;
7891 void *trace_data;
7892 int page_size;
7893 ssize_t ret = 0;
7894 ssize_t size;
7895
7896 if (!count)
7897 return 0;
7898
7899#ifdef CONFIG_TRACER_MAX_TRACE
7900 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7901 return -EBUSY;
7902#endif
7903
7904 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
7905
7906 /* Make sure the spare matches the current sub buffer size */
7907 if (info->spare) {
7908 if (page_size != info->spare_size) {
7909 ring_buffer_free_read_page(iter->array_buffer->buffer,
7910 info->spare_cpu, info->spare);
7911 info->spare = NULL;
7912 }
7913 }
7914
7915 if (!info->spare) {
7916 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7917 iter->cpu_file);
7918 if (IS_ERR(info->spare)) {
7919 ret = PTR_ERR(info->spare);
7920 info->spare = NULL;
7921 } else {
7922 info->spare_cpu = iter->cpu_file;
7923 info->spare_size = page_size;
7924 }
7925 }
7926 if (!info->spare)
7927 return ret;
7928
7929 /* Do we have previous read data to read? */
7930 if (info->read < page_size)
7931 goto read;
7932
7933 again:
7934 trace_access_lock(iter->cpu_file);
7935 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7936 info->spare,
7937 count,
7938 iter->cpu_file, 0);
7939 trace_access_unlock(iter->cpu_file);
7940
7941 if (ret < 0) {
7942 if (trace_empty(iter) && !iter->closed) {
7943 if ((filp->f_flags & O_NONBLOCK))
7944 return -EAGAIN;
7945
7946 ret = wait_on_pipe(iter, 0);
7947 if (ret)
7948 return ret;
7949
7950 goto again;
7951 }
7952 return 0;
7953 }
7954
7955 info->read = 0;
7956 read:
7957 size = page_size - info->read;
7958 if (size > count)
7959 size = count;
7960 trace_data = ring_buffer_read_page_data(info->spare);
7961 ret = copy_to_user(ubuf, trace_data + info->read, size);
7962 if (ret == size)
7963 return -EFAULT;
7964
7965 size -= ret;
7966
7967 *ppos += size;
7968 info->read += size;
7969
7970 return size;
7971}
7972
7973static int tracing_buffers_flush(struct file *file, fl_owner_t id)
7974{
7975 struct ftrace_buffer_info *info = file->private_data;
7976 struct trace_iterator *iter = &info->iter;
7977
7978 iter->closed = true;
7979 /* Make sure the waiters see the new wait_index */
7980 (void)atomic_fetch_inc_release(&iter->wait_index);
7981
7982 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
7983
7984 return 0;
7985}
7986
7987static int tracing_buffers_release(struct inode *inode, struct file *file)
7988{
7989 struct ftrace_buffer_info *info = file->private_data;
7990 struct trace_iterator *iter = &info->iter;
7991
7992 mutex_lock(&trace_types_lock);
7993
7994 iter->tr->trace_ref--;
7995
7996 __trace_array_put(iter->tr);
7997
7998 if (info->spare)
7999 ring_buffer_free_read_page(iter->array_buffer->buffer,
8000 info->spare_cpu, info->spare);
8001 kvfree(info);
8002
8003 mutex_unlock(&trace_types_lock);
8004
8005 return 0;
8006}
8007
8008struct buffer_ref {
8009 struct trace_buffer *buffer;
8010 void *page;
8011 int cpu;
8012 refcount_t refcount;
8013};
8014
8015static void buffer_ref_release(struct buffer_ref *ref)
8016{
8017 if (!refcount_dec_and_test(&ref->refcount))
8018 return;
8019 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8020 kfree(ref);
8021}
8022
8023static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8024 struct pipe_buffer *buf)
8025{
8026 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8027
8028 buffer_ref_release(ref);
8029 buf->private = 0;
8030}
8031
8032static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8033 struct pipe_buffer *buf)
8034{
8035 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8036
8037 if (refcount_read(&ref->refcount) > INT_MAX/2)
8038 return false;
8039
8040 refcount_inc(&ref->refcount);
8041 return true;
8042}
8043
8044/* Pipe buffer operations for a buffer. */
8045static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8046 .release = buffer_pipe_buf_release,
8047 .get = buffer_pipe_buf_get,
8048};
8049
8050/*
8051 * Callback from splice_to_pipe(), if we need to release some pages
8052 * at the end of the spd in case we error'ed out in filling the pipe.
8053 */
8054static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8055{
8056 struct buffer_ref *ref =
8057 (struct buffer_ref *)spd->partial[i].private;
8058
8059 buffer_ref_release(ref);
8060 spd->partial[i].private = 0;
8061}
8062
8063static ssize_t
8064tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8065 struct pipe_inode_info *pipe, size_t len,
8066 unsigned int flags)
8067{
8068 struct ftrace_buffer_info *info = file->private_data;
8069 struct trace_iterator *iter = &info->iter;
8070 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8071 struct page *pages_def[PIPE_DEF_BUFFERS];
8072 struct splice_pipe_desc spd = {
8073 .pages = pages_def,
8074 .partial = partial_def,
8075 .nr_pages_max = PIPE_DEF_BUFFERS,
8076 .ops = &buffer_pipe_buf_ops,
8077 .spd_release = buffer_spd_release,
8078 };
8079 struct buffer_ref *ref;
8080 bool woken = false;
8081 int page_size;
8082 int entries, i;
8083 ssize_t ret = 0;
8084
8085#ifdef CONFIG_TRACER_MAX_TRACE
8086 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8087 return -EBUSY;
8088#endif
8089
8090 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8091 if (*ppos & (page_size - 1))
8092 return -EINVAL;
8093
8094 if (len & (page_size - 1)) {
8095 if (len < page_size)
8096 return -EINVAL;
8097 len &= (~(page_size - 1));
8098 }
8099
8100 if (splice_grow_spd(pipe, &spd))
8101 return -ENOMEM;
8102
8103 again:
8104 trace_access_lock(iter->cpu_file);
8105 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8106
8107 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8108 struct page *page;
8109 int r;
8110
8111 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8112 if (!ref) {
8113 ret = -ENOMEM;
8114 break;
8115 }
8116
8117 refcount_set(&ref->refcount, 1);
8118 ref->buffer = iter->array_buffer->buffer;
8119 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8120 if (IS_ERR(ref->page)) {
8121 ret = PTR_ERR(ref->page);
8122 ref->page = NULL;
8123 kfree(ref);
8124 break;
8125 }
8126 ref->cpu = iter->cpu_file;
8127
8128 r = ring_buffer_read_page(ref->buffer, ref->page,
8129 len, iter->cpu_file, 1);
8130 if (r < 0) {
8131 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8132 ref->page);
8133 kfree(ref);
8134 break;
8135 }
8136
8137 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8138
8139 spd.pages[i] = page;
8140 spd.partial[i].len = page_size;
8141 spd.partial[i].offset = 0;
8142 spd.partial[i].private = (unsigned long)ref;
8143 spd.nr_pages++;
8144 *ppos += page_size;
8145
8146 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8147 }
8148
8149 trace_access_unlock(iter->cpu_file);
8150 spd.nr_pages = i;
8151
8152 /* did we read anything? */
8153 if (!spd.nr_pages) {
8154
8155 if (ret)
8156 goto out;
8157
8158 if (woken)
8159 goto out;
8160
8161 ret = -EAGAIN;
8162 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8163 goto out;
8164
8165 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8166 if (ret)
8167 goto out;
8168
8169 /* No need to wait after waking up when tracing is off */
8170 if (!tracer_tracing_is_on(iter->tr))
8171 goto out;
8172
8173 /* Iterate one more time to collect any new data then exit */
8174 woken = true;
8175
8176 goto again;
8177 }
8178
8179 ret = splice_to_pipe(pipe, &spd);
8180out:
8181 splice_shrink_spd(&spd);
8182
8183 return ret;
8184}
8185
8186static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8187{
8188 struct ftrace_buffer_info *info = file->private_data;
8189 struct trace_iterator *iter = &info->iter;
8190 int err;
8191
8192 if (cmd == TRACE_MMAP_IOCTL_GET_READER) {
8193 if (!(file->f_flags & O_NONBLOCK)) {
8194 err = ring_buffer_wait(iter->array_buffer->buffer,
8195 iter->cpu_file,
8196 iter->tr->buffer_percent,
8197 NULL, NULL);
8198 if (err)
8199 return err;
8200 }
8201
8202 return ring_buffer_map_get_reader(iter->array_buffer->buffer,
8203 iter->cpu_file);
8204 } else if (cmd) {
8205 return -ENOTTY;
8206 }
8207
8208 /*
8209 * An ioctl call with cmd 0 to the ring buffer file will wake up all
8210 * waiters
8211 */
8212 mutex_lock(&trace_types_lock);
8213
8214 /* Make sure the waiters see the new wait_index */
8215 (void)atomic_fetch_inc_release(&iter->wait_index);
8216
8217 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8218
8219 mutex_unlock(&trace_types_lock);
8220 return 0;
8221}
8222
8223#ifdef CONFIG_TRACER_MAX_TRACE
8224static int get_snapshot_map(struct trace_array *tr)
8225{
8226 int err = 0;
8227
8228 /*
8229 * Called with mmap_lock held. lockdep would be unhappy if we would now
8230 * take trace_types_lock. Instead use the specific
8231 * snapshot_trigger_lock.
8232 */
8233 spin_lock(&tr->snapshot_trigger_lock);
8234
8235 if (tr->snapshot || tr->mapped == UINT_MAX)
8236 err = -EBUSY;
8237 else
8238 tr->mapped++;
8239
8240 spin_unlock(&tr->snapshot_trigger_lock);
8241
8242 /* Wait for update_max_tr() to observe iter->tr->mapped */
8243 if (tr->mapped == 1)
8244 synchronize_rcu();
8245
8246 return err;
8247
8248}
8249static void put_snapshot_map(struct trace_array *tr)
8250{
8251 spin_lock(&tr->snapshot_trigger_lock);
8252 if (!WARN_ON(!tr->mapped))
8253 tr->mapped--;
8254 spin_unlock(&tr->snapshot_trigger_lock);
8255}
8256#else
8257static inline int get_snapshot_map(struct trace_array *tr) { return 0; }
8258static inline void put_snapshot_map(struct trace_array *tr) { }
8259#endif
8260
8261static void tracing_buffers_mmap_close(struct vm_area_struct *vma)
8262{
8263 struct ftrace_buffer_info *info = vma->vm_file->private_data;
8264 struct trace_iterator *iter = &info->iter;
8265
8266 WARN_ON(ring_buffer_unmap(iter->array_buffer->buffer, iter->cpu_file));
8267 put_snapshot_map(iter->tr);
8268}
8269
8270static const struct vm_operations_struct tracing_buffers_vmops = {
8271 .close = tracing_buffers_mmap_close,
8272};
8273
8274static int tracing_buffers_mmap(struct file *filp, struct vm_area_struct *vma)
8275{
8276 struct ftrace_buffer_info *info = filp->private_data;
8277 struct trace_iterator *iter = &info->iter;
8278 int ret = 0;
8279
8280 /* Currently the boot mapped buffer is not supported for mmap */
8281 if (iter->tr->flags & TRACE_ARRAY_FL_BOOT)
8282 return -ENODEV;
8283
8284 ret = get_snapshot_map(iter->tr);
8285 if (ret)
8286 return ret;
8287
8288 ret = ring_buffer_map(iter->array_buffer->buffer, iter->cpu_file, vma);
8289 if (ret)
8290 put_snapshot_map(iter->tr);
8291
8292 vma->vm_ops = &tracing_buffers_vmops;
8293
8294 return ret;
8295}
8296
8297static const struct file_operations tracing_buffers_fops = {
8298 .open = tracing_buffers_open,
8299 .read = tracing_buffers_read,
8300 .poll = tracing_buffers_poll,
8301 .release = tracing_buffers_release,
8302 .flush = tracing_buffers_flush,
8303 .splice_read = tracing_buffers_splice_read,
8304 .unlocked_ioctl = tracing_buffers_ioctl,
8305 .mmap = tracing_buffers_mmap,
8306};
8307
8308static ssize_t
8309tracing_stats_read(struct file *filp, char __user *ubuf,
8310 size_t count, loff_t *ppos)
8311{
8312 struct inode *inode = file_inode(filp);
8313 struct trace_array *tr = inode->i_private;
8314 struct array_buffer *trace_buf = &tr->array_buffer;
8315 int cpu = tracing_get_cpu(inode);
8316 struct trace_seq *s;
8317 unsigned long cnt;
8318 unsigned long long t;
8319 unsigned long usec_rem;
8320
8321 s = kmalloc(sizeof(*s), GFP_KERNEL);
8322 if (!s)
8323 return -ENOMEM;
8324
8325 trace_seq_init(s);
8326
8327 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8328 trace_seq_printf(s, "entries: %ld\n", cnt);
8329
8330 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8331 trace_seq_printf(s, "overrun: %ld\n", cnt);
8332
8333 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8334 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8335
8336 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8337 trace_seq_printf(s, "bytes: %ld\n", cnt);
8338
8339 if (trace_clocks[tr->clock_id].in_ns) {
8340 /* local or global for trace_clock */
8341 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8342 usec_rem = do_div(t, USEC_PER_SEC);
8343 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8344 t, usec_rem);
8345
8346 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8347 usec_rem = do_div(t, USEC_PER_SEC);
8348 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8349 } else {
8350 /* counter or tsc mode for trace_clock */
8351 trace_seq_printf(s, "oldest event ts: %llu\n",
8352 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8353
8354 trace_seq_printf(s, "now ts: %llu\n",
8355 ring_buffer_time_stamp(trace_buf->buffer));
8356 }
8357
8358 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8359 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8360
8361 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8362 trace_seq_printf(s, "read events: %ld\n", cnt);
8363
8364 count = simple_read_from_buffer(ubuf, count, ppos,
8365 s->buffer, trace_seq_used(s));
8366
8367 kfree(s);
8368
8369 return count;
8370}
8371
8372static const struct file_operations tracing_stats_fops = {
8373 .open = tracing_open_generic_tr,
8374 .read = tracing_stats_read,
8375 .llseek = generic_file_llseek,
8376 .release = tracing_release_generic_tr,
8377};
8378
8379#ifdef CONFIG_DYNAMIC_FTRACE
8380
8381static ssize_t
8382tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8383 size_t cnt, loff_t *ppos)
8384{
8385 ssize_t ret;
8386 char *buf;
8387 int r;
8388
8389 /* 512 should be plenty to hold the amount needed */
8390#define DYN_INFO_BUF_SIZE 512
8391
8392 buf = kmalloc(DYN_INFO_BUF_SIZE, GFP_KERNEL);
8393 if (!buf)
8394 return -ENOMEM;
8395
8396 r = scnprintf(buf, DYN_INFO_BUF_SIZE,
8397 "%ld pages:%ld groups: %ld\n"
8398 "ftrace boot update time = %llu (ns)\n"
8399 "ftrace module total update time = %llu (ns)\n",
8400 ftrace_update_tot_cnt,
8401 ftrace_number_of_pages,
8402 ftrace_number_of_groups,
8403 ftrace_update_time,
8404 ftrace_total_mod_time);
8405
8406 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8407 kfree(buf);
8408 return ret;
8409}
8410
8411static const struct file_operations tracing_dyn_info_fops = {
8412 .open = tracing_open_generic,
8413 .read = tracing_read_dyn_info,
8414 .llseek = generic_file_llseek,
8415};
8416#endif /* CONFIG_DYNAMIC_FTRACE */
8417
8418#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8419static void
8420ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8421 struct trace_array *tr, struct ftrace_probe_ops *ops,
8422 void *data)
8423{
8424 tracing_snapshot_instance(tr);
8425}
8426
8427static void
8428ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8429 struct trace_array *tr, struct ftrace_probe_ops *ops,
8430 void *data)
8431{
8432 struct ftrace_func_mapper *mapper = data;
8433 long *count = NULL;
8434
8435 if (mapper)
8436 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8437
8438 if (count) {
8439
8440 if (*count <= 0)
8441 return;
8442
8443 (*count)--;
8444 }
8445
8446 tracing_snapshot_instance(tr);
8447}
8448
8449static int
8450ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8451 struct ftrace_probe_ops *ops, void *data)
8452{
8453 struct ftrace_func_mapper *mapper = data;
8454 long *count = NULL;
8455
8456 seq_printf(m, "%ps:", (void *)ip);
8457
8458 seq_puts(m, "snapshot");
8459
8460 if (mapper)
8461 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8462
8463 if (count)
8464 seq_printf(m, ":count=%ld\n", *count);
8465 else
8466 seq_puts(m, ":unlimited\n");
8467
8468 return 0;
8469}
8470
8471static int
8472ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8473 unsigned long ip, void *init_data, void **data)
8474{
8475 struct ftrace_func_mapper *mapper = *data;
8476
8477 if (!mapper) {
8478 mapper = allocate_ftrace_func_mapper();
8479 if (!mapper)
8480 return -ENOMEM;
8481 *data = mapper;
8482 }
8483
8484 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8485}
8486
8487static void
8488ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8489 unsigned long ip, void *data)
8490{
8491 struct ftrace_func_mapper *mapper = data;
8492
8493 if (!ip) {
8494 if (!mapper)
8495 return;
8496 free_ftrace_func_mapper(mapper, NULL);
8497 return;
8498 }
8499
8500 ftrace_func_mapper_remove_ip(mapper, ip);
8501}
8502
8503static struct ftrace_probe_ops snapshot_probe_ops = {
8504 .func = ftrace_snapshot,
8505 .print = ftrace_snapshot_print,
8506};
8507
8508static struct ftrace_probe_ops snapshot_count_probe_ops = {
8509 .func = ftrace_count_snapshot,
8510 .print = ftrace_snapshot_print,
8511 .init = ftrace_snapshot_init,
8512 .free = ftrace_snapshot_free,
8513};
8514
8515static int
8516ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8517 char *glob, char *cmd, char *param, int enable)
8518{
8519 struct ftrace_probe_ops *ops;
8520 void *count = (void *)-1;
8521 char *number;
8522 int ret;
8523
8524 if (!tr)
8525 return -ENODEV;
8526
8527 /* hash funcs only work with set_ftrace_filter */
8528 if (!enable)
8529 return -EINVAL;
8530
8531 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8532
8533 if (glob[0] == '!') {
8534 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
8535 if (!ret)
8536 tracing_disarm_snapshot(tr);
8537
8538 return ret;
8539 }
8540
8541 if (!param)
8542 goto out_reg;
8543
8544 number = strsep(¶m, ":");
8545
8546 if (!strlen(number))
8547 goto out_reg;
8548
8549 /*
8550 * We use the callback data field (which is a pointer)
8551 * as our counter.
8552 */
8553 ret = kstrtoul(number, 0, (unsigned long *)&count);
8554 if (ret)
8555 return ret;
8556
8557 out_reg:
8558 ret = tracing_arm_snapshot(tr);
8559 if (ret < 0)
8560 goto out;
8561
8562 ret = register_ftrace_function_probe(glob, tr, ops, count);
8563 if (ret < 0)
8564 tracing_disarm_snapshot(tr);
8565 out:
8566 return ret < 0 ? ret : 0;
8567}
8568
8569static struct ftrace_func_command ftrace_snapshot_cmd = {
8570 .name = "snapshot",
8571 .func = ftrace_trace_snapshot_callback,
8572};
8573
8574static __init int register_snapshot_cmd(void)
8575{
8576 return register_ftrace_command(&ftrace_snapshot_cmd);
8577}
8578#else
8579static inline __init int register_snapshot_cmd(void) { return 0; }
8580#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8581
8582static struct dentry *tracing_get_dentry(struct trace_array *tr)
8583{
8584 if (WARN_ON(!tr->dir))
8585 return ERR_PTR(-ENODEV);
8586
8587 /* Top directory uses NULL as the parent */
8588 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8589 return NULL;
8590
8591 /* All sub buffers have a descriptor */
8592 return tr->dir;
8593}
8594
8595static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8596{
8597 struct dentry *d_tracer;
8598
8599 if (tr->percpu_dir)
8600 return tr->percpu_dir;
8601
8602 d_tracer = tracing_get_dentry(tr);
8603 if (IS_ERR(d_tracer))
8604 return NULL;
8605
8606 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8607
8608 MEM_FAIL(!tr->percpu_dir,
8609 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8610
8611 return tr->percpu_dir;
8612}
8613
8614static struct dentry *
8615trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8616 void *data, long cpu, const struct file_operations *fops)
8617{
8618 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8619
8620 if (ret) /* See tracing_get_cpu() */
8621 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8622 return ret;
8623}
8624
8625static void
8626tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8627{
8628 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8629 struct dentry *d_cpu;
8630 char cpu_dir[30]; /* 30 characters should be more than enough */
8631
8632 if (!d_percpu)
8633 return;
8634
8635 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8636 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8637 if (!d_cpu) {
8638 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8639 return;
8640 }
8641
8642 /* per cpu trace_pipe */
8643 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8644 tr, cpu, &tracing_pipe_fops);
8645
8646 /* per cpu trace */
8647 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8648 tr, cpu, &tracing_fops);
8649
8650 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8651 tr, cpu, &tracing_buffers_fops);
8652
8653 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8654 tr, cpu, &tracing_stats_fops);
8655
8656 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8657 tr, cpu, &tracing_entries_fops);
8658
8659 if (tr->range_addr_start)
8660 trace_create_cpu_file("buffer_meta", TRACE_MODE_READ, d_cpu,
8661 tr, cpu, &tracing_buffer_meta_fops);
8662#ifdef CONFIG_TRACER_SNAPSHOT
8663 if (!tr->range_addr_start) {
8664 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8665 tr, cpu, &snapshot_fops);
8666
8667 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8668 tr, cpu, &snapshot_raw_fops);
8669 }
8670#endif
8671}
8672
8673#ifdef CONFIG_FTRACE_SELFTEST
8674/* Let selftest have access to static functions in this file */
8675#include "trace_selftest.c"
8676#endif
8677
8678static ssize_t
8679trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8680 loff_t *ppos)
8681{
8682 struct trace_option_dentry *topt = filp->private_data;
8683 char *buf;
8684
8685 if (topt->flags->val & topt->opt->bit)
8686 buf = "1\n";
8687 else
8688 buf = "0\n";
8689
8690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8691}
8692
8693static ssize_t
8694trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8695 loff_t *ppos)
8696{
8697 struct trace_option_dentry *topt = filp->private_data;
8698 unsigned long val;
8699 int ret;
8700
8701 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8702 if (ret)
8703 return ret;
8704
8705 if (val != 0 && val != 1)
8706 return -EINVAL;
8707
8708 if (!!(topt->flags->val & topt->opt->bit) != val) {
8709 mutex_lock(&trace_types_lock);
8710 ret = __set_tracer_option(topt->tr, topt->flags,
8711 topt->opt, !val);
8712 mutex_unlock(&trace_types_lock);
8713 if (ret)
8714 return ret;
8715 }
8716
8717 *ppos += cnt;
8718
8719 return cnt;
8720}
8721
8722static int tracing_open_options(struct inode *inode, struct file *filp)
8723{
8724 struct trace_option_dentry *topt = inode->i_private;
8725 int ret;
8726
8727 ret = tracing_check_open_get_tr(topt->tr);
8728 if (ret)
8729 return ret;
8730
8731 filp->private_data = inode->i_private;
8732 return 0;
8733}
8734
8735static int tracing_release_options(struct inode *inode, struct file *file)
8736{
8737 struct trace_option_dentry *topt = file->private_data;
8738
8739 trace_array_put(topt->tr);
8740 return 0;
8741}
8742
8743static const struct file_operations trace_options_fops = {
8744 .open = tracing_open_options,
8745 .read = trace_options_read,
8746 .write = trace_options_write,
8747 .llseek = generic_file_llseek,
8748 .release = tracing_release_options,
8749};
8750
8751/*
8752 * In order to pass in both the trace_array descriptor as well as the index
8753 * to the flag that the trace option file represents, the trace_array
8754 * has a character array of trace_flags_index[], which holds the index
8755 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8756 * The address of this character array is passed to the flag option file
8757 * read/write callbacks.
8758 *
8759 * In order to extract both the index and the trace_array descriptor,
8760 * get_tr_index() uses the following algorithm.
8761 *
8762 * idx = *ptr;
8763 *
8764 * As the pointer itself contains the address of the index (remember
8765 * index[1] == 1).
8766 *
8767 * Then to get the trace_array descriptor, by subtracting that index
8768 * from the ptr, we get to the start of the index itself.
8769 *
8770 * ptr - idx == &index[0]
8771 *
8772 * Then a simple container_of() from that pointer gets us to the
8773 * trace_array descriptor.
8774 */
8775static void get_tr_index(void *data, struct trace_array **ptr,
8776 unsigned int *pindex)
8777{
8778 *pindex = *(unsigned char *)data;
8779
8780 *ptr = container_of(data - *pindex, struct trace_array,
8781 trace_flags_index);
8782}
8783
8784static ssize_t
8785trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8786 loff_t *ppos)
8787{
8788 void *tr_index = filp->private_data;
8789 struct trace_array *tr;
8790 unsigned int index;
8791 char *buf;
8792
8793 get_tr_index(tr_index, &tr, &index);
8794
8795 if (tr->trace_flags & (1 << index))
8796 buf = "1\n";
8797 else
8798 buf = "0\n";
8799
8800 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8801}
8802
8803static ssize_t
8804trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8805 loff_t *ppos)
8806{
8807 void *tr_index = filp->private_data;
8808 struct trace_array *tr;
8809 unsigned int index;
8810 unsigned long val;
8811 int ret;
8812
8813 get_tr_index(tr_index, &tr, &index);
8814
8815 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8816 if (ret)
8817 return ret;
8818
8819 if (val != 0 && val != 1)
8820 return -EINVAL;
8821
8822 mutex_lock(&event_mutex);
8823 mutex_lock(&trace_types_lock);
8824 ret = set_tracer_flag(tr, 1 << index, val);
8825 mutex_unlock(&trace_types_lock);
8826 mutex_unlock(&event_mutex);
8827
8828 if (ret < 0)
8829 return ret;
8830
8831 *ppos += cnt;
8832
8833 return cnt;
8834}
8835
8836static const struct file_operations trace_options_core_fops = {
8837 .open = tracing_open_generic,
8838 .read = trace_options_core_read,
8839 .write = trace_options_core_write,
8840 .llseek = generic_file_llseek,
8841};
8842
8843struct dentry *trace_create_file(const char *name,
8844 umode_t mode,
8845 struct dentry *parent,
8846 void *data,
8847 const struct file_operations *fops)
8848{
8849 struct dentry *ret;
8850
8851 ret = tracefs_create_file(name, mode, parent, data, fops);
8852 if (!ret)
8853 pr_warn("Could not create tracefs '%s' entry\n", name);
8854
8855 return ret;
8856}
8857
8858
8859static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8860{
8861 struct dentry *d_tracer;
8862
8863 if (tr->options)
8864 return tr->options;
8865
8866 d_tracer = tracing_get_dentry(tr);
8867 if (IS_ERR(d_tracer))
8868 return NULL;
8869
8870 tr->options = tracefs_create_dir("options", d_tracer);
8871 if (!tr->options) {
8872 pr_warn("Could not create tracefs directory 'options'\n");
8873 return NULL;
8874 }
8875
8876 return tr->options;
8877}
8878
8879static void
8880create_trace_option_file(struct trace_array *tr,
8881 struct trace_option_dentry *topt,
8882 struct tracer_flags *flags,
8883 struct tracer_opt *opt)
8884{
8885 struct dentry *t_options;
8886
8887 t_options = trace_options_init_dentry(tr);
8888 if (!t_options)
8889 return;
8890
8891 topt->flags = flags;
8892 topt->opt = opt;
8893 topt->tr = tr;
8894
8895 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8896 t_options, topt, &trace_options_fops);
8897
8898}
8899
8900static void
8901create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8902{
8903 struct trace_option_dentry *topts;
8904 struct trace_options *tr_topts;
8905 struct tracer_flags *flags;
8906 struct tracer_opt *opts;
8907 int cnt;
8908 int i;
8909
8910 if (!tracer)
8911 return;
8912
8913 flags = tracer->flags;
8914
8915 if (!flags || !flags->opts)
8916 return;
8917
8918 /*
8919 * If this is an instance, only create flags for tracers
8920 * the instance may have.
8921 */
8922 if (!trace_ok_for_array(tracer, tr))
8923 return;
8924
8925 for (i = 0; i < tr->nr_topts; i++) {
8926 /* Make sure there's no duplicate flags. */
8927 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8928 return;
8929 }
8930
8931 opts = flags->opts;
8932
8933 for (cnt = 0; opts[cnt].name; cnt++)
8934 ;
8935
8936 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8937 if (!topts)
8938 return;
8939
8940 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8941 GFP_KERNEL);
8942 if (!tr_topts) {
8943 kfree(topts);
8944 return;
8945 }
8946
8947 tr->topts = tr_topts;
8948 tr->topts[tr->nr_topts].tracer = tracer;
8949 tr->topts[tr->nr_topts].topts = topts;
8950 tr->nr_topts++;
8951
8952 for (cnt = 0; opts[cnt].name; cnt++) {
8953 create_trace_option_file(tr, &topts[cnt], flags,
8954 &opts[cnt]);
8955 MEM_FAIL(topts[cnt].entry == NULL,
8956 "Failed to create trace option: %s",
8957 opts[cnt].name);
8958 }
8959}
8960
8961static struct dentry *
8962create_trace_option_core_file(struct trace_array *tr,
8963 const char *option, long index)
8964{
8965 struct dentry *t_options;
8966
8967 t_options = trace_options_init_dentry(tr);
8968 if (!t_options)
8969 return NULL;
8970
8971 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
8972 (void *)&tr->trace_flags_index[index],
8973 &trace_options_core_fops);
8974}
8975
8976static void create_trace_options_dir(struct trace_array *tr)
8977{
8978 struct dentry *t_options;
8979 bool top_level = tr == &global_trace;
8980 int i;
8981
8982 t_options = trace_options_init_dentry(tr);
8983 if (!t_options)
8984 return;
8985
8986 for (i = 0; trace_options[i]; i++) {
8987 if (top_level ||
8988 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8989 create_trace_option_core_file(tr, trace_options[i], i);
8990 }
8991}
8992
8993static ssize_t
8994rb_simple_read(struct file *filp, char __user *ubuf,
8995 size_t cnt, loff_t *ppos)
8996{
8997 struct trace_array *tr = filp->private_data;
8998 char buf[64];
8999 int r;
9000
9001 r = tracer_tracing_is_on(tr);
9002 r = sprintf(buf, "%d\n", r);
9003
9004 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9005}
9006
9007static ssize_t
9008rb_simple_write(struct file *filp, const char __user *ubuf,
9009 size_t cnt, loff_t *ppos)
9010{
9011 struct trace_array *tr = filp->private_data;
9012 struct trace_buffer *buffer = tr->array_buffer.buffer;
9013 unsigned long val;
9014 int ret;
9015
9016 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9017 if (ret)
9018 return ret;
9019
9020 if (buffer) {
9021 mutex_lock(&trace_types_lock);
9022 if (!!val == tracer_tracing_is_on(tr)) {
9023 val = 0; /* do nothing */
9024 } else if (val) {
9025 tracer_tracing_on(tr);
9026 if (tr->current_trace->start)
9027 tr->current_trace->start(tr);
9028 } else {
9029 tracer_tracing_off(tr);
9030 if (tr->current_trace->stop)
9031 tr->current_trace->stop(tr);
9032 /* Wake up any waiters */
9033 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9034 }
9035 mutex_unlock(&trace_types_lock);
9036 }
9037
9038 (*ppos)++;
9039
9040 return cnt;
9041}
9042
9043static const struct file_operations rb_simple_fops = {
9044 .open = tracing_open_generic_tr,
9045 .read = rb_simple_read,
9046 .write = rb_simple_write,
9047 .release = tracing_release_generic_tr,
9048 .llseek = default_llseek,
9049};
9050
9051static ssize_t
9052buffer_percent_read(struct file *filp, char __user *ubuf,
9053 size_t cnt, loff_t *ppos)
9054{
9055 struct trace_array *tr = filp->private_data;
9056 char buf[64];
9057 int r;
9058
9059 r = tr->buffer_percent;
9060 r = sprintf(buf, "%d\n", r);
9061
9062 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9063}
9064
9065static ssize_t
9066buffer_percent_write(struct file *filp, const char __user *ubuf,
9067 size_t cnt, loff_t *ppos)
9068{
9069 struct trace_array *tr = filp->private_data;
9070 unsigned long val;
9071 int ret;
9072
9073 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9074 if (ret)
9075 return ret;
9076
9077 if (val > 100)
9078 return -EINVAL;
9079
9080 tr->buffer_percent = val;
9081
9082 (*ppos)++;
9083
9084 return cnt;
9085}
9086
9087static const struct file_operations buffer_percent_fops = {
9088 .open = tracing_open_generic_tr,
9089 .read = buffer_percent_read,
9090 .write = buffer_percent_write,
9091 .release = tracing_release_generic_tr,
9092 .llseek = default_llseek,
9093};
9094
9095static ssize_t
9096buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9097{
9098 struct trace_array *tr = filp->private_data;
9099 size_t size;
9100 char buf[64];
9101 int order;
9102 int r;
9103
9104 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9105 size = (PAGE_SIZE << order) / 1024;
9106
9107 r = sprintf(buf, "%zd\n", size);
9108
9109 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9110}
9111
9112static ssize_t
9113buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9114 size_t cnt, loff_t *ppos)
9115{
9116 struct trace_array *tr = filp->private_data;
9117 unsigned long val;
9118 int old_order;
9119 int order;
9120 int pages;
9121 int ret;
9122
9123 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9124 if (ret)
9125 return ret;
9126
9127 val *= 1024; /* value passed in is in KB */
9128
9129 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9130 order = fls(pages - 1);
9131
9132 /* limit between 1 and 128 system pages */
9133 if (order < 0 || order > 7)
9134 return -EINVAL;
9135
9136 /* Do not allow tracing while changing the order of the ring buffer */
9137 tracing_stop_tr(tr);
9138
9139 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9140 if (old_order == order)
9141 goto out;
9142
9143 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9144 if (ret)
9145 goto out;
9146
9147#ifdef CONFIG_TRACER_MAX_TRACE
9148
9149 if (!tr->allocated_snapshot)
9150 goto out_max;
9151
9152 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9153 if (ret) {
9154 /* Put back the old order */
9155 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9156 if (WARN_ON_ONCE(cnt)) {
9157 /*
9158 * AARGH! We are left with different orders!
9159 * The max buffer is our "snapshot" buffer.
9160 * When a tracer needs a snapshot (one of the
9161 * latency tracers), it swaps the max buffer
9162 * with the saved snap shot. We succeeded to
9163 * update the order of the main buffer, but failed to
9164 * update the order of the max buffer. But when we tried
9165 * to reset the main buffer to the original size, we
9166 * failed there too. This is very unlikely to
9167 * happen, but if it does, warn and kill all
9168 * tracing.
9169 */
9170 tracing_disabled = 1;
9171 }
9172 goto out;
9173 }
9174 out_max:
9175#endif
9176 (*ppos)++;
9177 out:
9178 if (ret)
9179 cnt = ret;
9180 tracing_start_tr(tr);
9181 return cnt;
9182}
9183
9184static const struct file_operations buffer_subbuf_size_fops = {
9185 .open = tracing_open_generic_tr,
9186 .read = buffer_subbuf_size_read,
9187 .write = buffer_subbuf_size_write,
9188 .release = tracing_release_generic_tr,
9189 .llseek = default_llseek,
9190};
9191
9192static struct dentry *trace_instance_dir;
9193
9194static void
9195init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9196
9197static int
9198allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9199{
9200 enum ring_buffer_flags rb_flags;
9201
9202 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9203
9204 buf->tr = tr;
9205
9206 if (tr->range_addr_start && tr->range_addr_size) {
9207 buf->buffer = ring_buffer_alloc_range(size, rb_flags, 0,
9208 tr->range_addr_start,
9209 tr->range_addr_size);
9210
9211 ring_buffer_last_boot_delta(buf->buffer,
9212 &tr->text_delta, &tr->data_delta);
9213 /*
9214 * This is basically the same as a mapped buffer,
9215 * with the same restrictions.
9216 */
9217 tr->mapped++;
9218 } else {
9219 buf->buffer = ring_buffer_alloc(size, rb_flags);
9220 }
9221 if (!buf->buffer)
9222 return -ENOMEM;
9223
9224 buf->data = alloc_percpu(struct trace_array_cpu);
9225 if (!buf->data) {
9226 ring_buffer_free(buf->buffer);
9227 buf->buffer = NULL;
9228 return -ENOMEM;
9229 }
9230
9231 /* Allocate the first page for all buffers */
9232 set_buffer_entries(&tr->array_buffer,
9233 ring_buffer_size(tr->array_buffer.buffer, 0));
9234
9235 return 0;
9236}
9237
9238static void free_trace_buffer(struct array_buffer *buf)
9239{
9240 if (buf->buffer) {
9241 ring_buffer_free(buf->buffer);
9242 buf->buffer = NULL;
9243 free_percpu(buf->data);
9244 buf->data = NULL;
9245 }
9246}
9247
9248static int allocate_trace_buffers(struct trace_array *tr, int size)
9249{
9250 int ret;
9251
9252 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9253 if (ret)
9254 return ret;
9255
9256#ifdef CONFIG_TRACER_MAX_TRACE
9257 /* Fix mapped buffer trace arrays do not have snapshot buffers */
9258 if (tr->range_addr_start)
9259 return 0;
9260
9261 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9262 allocate_snapshot ? size : 1);
9263 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9264 free_trace_buffer(&tr->array_buffer);
9265 return -ENOMEM;
9266 }
9267 tr->allocated_snapshot = allocate_snapshot;
9268
9269 allocate_snapshot = false;
9270#endif
9271
9272 return 0;
9273}
9274
9275static void free_trace_buffers(struct trace_array *tr)
9276{
9277 if (!tr)
9278 return;
9279
9280 free_trace_buffer(&tr->array_buffer);
9281
9282#ifdef CONFIG_TRACER_MAX_TRACE
9283 free_trace_buffer(&tr->max_buffer);
9284#endif
9285}
9286
9287static void init_trace_flags_index(struct trace_array *tr)
9288{
9289 int i;
9290
9291 /* Used by the trace options files */
9292 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9293 tr->trace_flags_index[i] = i;
9294}
9295
9296static void __update_tracer_options(struct trace_array *tr)
9297{
9298 struct tracer *t;
9299
9300 for (t = trace_types; t; t = t->next)
9301 add_tracer_options(tr, t);
9302}
9303
9304static void update_tracer_options(struct trace_array *tr)
9305{
9306 mutex_lock(&trace_types_lock);
9307 tracer_options_updated = true;
9308 __update_tracer_options(tr);
9309 mutex_unlock(&trace_types_lock);
9310}
9311
9312/* Must have trace_types_lock held */
9313struct trace_array *trace_array_find(const char *instance)
9314{
9315 struct trace_array *tr, *found = NULL;
9316
9317 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9318 if (tr->name && strcmp(tr->name, instance) == 0) {
9319 found = tr;
9320 break;
9321 }
9322 }
9323
9324 return found;
9325}
9326
9327struct trace_array *trace_array_find_get(const char *instance)
9328{
9329 struct trace_array *tr;
9330
9331 mutex_lock(&trace_types_lock);
9332 tr = trace_array_find(instance);
9333 if (tr)
9334 tr->ref++;
9335 mutex_unlock(&trace_types_lock);
9336
9337 return tr;
9338}
9339
9340static int trace_array_create_dir(struct trace_array *tr)
9341{
9342 int ret;
9343
9344 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9345 if (!tr->dir)
9346 return -EINVAL;
9347
9348 ret = event_trace_add_tracer(tr->dir, tr);
9349 if (ret) {
9350 tracefs_remove(tr->dir);
9351 return ret;
9352 }
9353
9354 init_tracer_tracefs(tr, tr->dir);
9355 __update_tracer_options(tr);
9356
9357 return ret;
9358}
9359
9360static struct trace_array *
9361trace_array_create_systems(const char *name, const char *systems,
9362 unsigned long range_addr_start,
9363 unsigned long range_addr_size)
9364{
9365 struct trace_array *tr;
9366 int ret;
9367
9368 ret = -ENOMEM;
9369 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9370 if (!tr)
9371 return ERR_PTR(ret);
9372
9373 tr->name = kstrdup(name, GFP_KERNEL);
9374 if (!tr->name)
9375 goto out_free_tr;
9376
9377 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9378 goto out_free_tr;
9379
9380 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9381 goto out_free_tr;
9382
9383 if (systems) {
9384 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9385 if (!tr->system_names)
9386 goto out_free_tr;
9387 }
9388
9389 /* Only for boot up memory mapped ring buffers */
9390 tr->range_addr_start = range_addr_start;
9391 tr->range_addr_size = range_addr_size;
9392
9393 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9394
9395 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9396
9397 raw_spin_lock_init(&tr->start_lock);
9398
9399 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9400#ifdef CONFIG_TRACER_MAX_TRACE
9401 spin_lock_init(&tr->snapshot_trigger_lock);
9402#endif
9403 tr->current_trace = &nop_trace;
9404
9405 INIT_LIST_HEAD(&tr->systems);
9406 INIT_LIST_HEAD(&tr->events);
9407 INIT_LIST_HEAD(&tr->hist_vars);
9408 INIT_LIST_HEAD(&tr->err_log);
9409
9410 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9411 goto out_free_tr;
9412
9413 /* The ring buffer is defaultly expanded */
9414 trace_set_ring_buffer_expanded(tr);
9415
9416 if (ftrace_allocate_ftrace_ops(tr) < 0)
9417 goto out_free_tr;
9418
9419 ftrace_init_trace_array(tr);
9420
9421 init_trace_flags_index(tr);
9422
9423 if (trace_instance_dir) {
9424 ret = trace_array_create_dir(tr);
9425 if (ret)
9426 goto out_free_tr;
9427 } else
9428 __trace_early_add_events(tr);
9429
9430 list_add(&tr->list, &ftrace_trace_arrays);
9431
9432 tr->ref++;
9433
9434 return tr;
9435
9436 out_free_tr:
9437 ftrace_free_ftrace_ops(tr);
9438 free_trace_buffers(tr);
9439 free_cpumask_var(tr->pipe_cpumask);
9440 free_cpumask_var(tr->tracing_cpumask);
9441 kfree_const(tr->system_names);
9442 kfree(tr->name);
9443 kfree(tr);
9444
9445 return ERR_PTR(ret);
9446}
9447
9448static struct trace_array *trace_array_create(const char *name)
9449{
9450 return trace_array_create_systems(name, NULL, 0, 0);
9451}
9452
9453static int instance_mkdir(const char *name)
9454{
9455 struct trace_array *tr;
9456 int ret;
9457
9458 guard(mutex)(&event_mutex);
9459 guard(mutex)(&trace_types_lock);
9460
9461 ret = -EEXIST;
9462 if (trace_array_find(name))
9463 return -EEXIST;
9464
9465 tr = trace_array_create(name);
9466
9467 ret = PTR_ERR_OR_ZERO(tr);
9468
9469 return ret;
9470}
9471
9472static u64 map_pages(u64 start, u64 size)
9473{
9474 struct page **pages;
9475 phys_addr_t page_start;
9476 unsigned int page_count;
9477 unsigned int i;
9478 void *vaddr;
9479
9480 page_count = DIV_ROUND_UP(size, PAGE_SIZE);
9481
9482 page_start = start;
9483 pages = kmalloc_array(page_count, sizeof(struct page *), GFP_KERNEL);
9484 if (!pages)
9485 return 0;
9486
9487 for (i = 0; i < page_count; i++) {
9488 phys_addr_t addr = page_start + i * PAGE_SIZE;
9489 pages[i] = pfn_to_page(addr >> PAGE_SHIFT);
9490 }
9491 vaddr = vmap(pages, page_count, VM_MAP, PAGE_KERNEL);
9492 kfree(pages);
9493
9494 return (u64)(unsigned long)vaddr;
9495}
9496
9497/**
9498 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9499 * @name: The name of the trace array to be looked up/created.
9500 * @systems: A list of systems to create event directories for (NULL for all)
9501 *
9502 * Returns pointer to trace array with given name.
9503 * NULL, if it cannot be created.
9504 *
9505 * NOTE: This function increments the reference counter associated with the
9506 * trace array returned. This makes sure it cannot be freed while in use.
9507 * Use trace_array_put() once the trace array is no longer needed.
9508 * If the trace_array is to be freed, trace_array_destroy() needs to
9509 * be called after the trace_array_put(), or simply let user space delete
9510 * it from the tracefs instances directory. But until the
9511 * trace_array_put() is called, user space can not delete it.
9512 *
9513 */
9514struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9515{
9516 struct trace_array *tr;
9517
9518 guard(mutex)(&event_mutex);
9519 guard(mutex)(&trace_types_lock);
9520
9521 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9522 if (tr->name && strcmp(tr->name, name) == 0) {
9523 tr->ref++;
9524 return tr;
9525 }
9526 }
9527
9528 tr = trace_array_create_systems(name, systems, 0, 0);
9529
9530 if (IS_ERR(tr))
9531 tr = NULL;
9532 else
9533 tr->ref++;
9534
9535 return tr;
9536}
9537EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9538
9539static int __remove_instance(struct trace_array *tr)
9540{
9541 int i;
9542
9543 /* Reference counter for a newly created trace array = 1. */
9544 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9545 return -EBUSY;
9546
9547 list_del(&tr->list);
9548
9549 /* Disable all the flags that were enabled coming in */
9550 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9551 if ((1 << i) & ZEROED_TRACE_FLAGS)
9552 set_tracer_flag(tr, 1 << i, 0);
9553 }
9554
9555 if (printk_trace == tr)
9556 update_printk_trace(&global_trace);
9557
9558 tracing_set_nop(tr);
9559 clear_ftrace_function_probes(tr);
9560 event_trace_del_tracer(tr);
9561 ftrace_clear_pids(tr);
9562 ftrace_destroy_function_files(tr);
9563 tracefs_remove(tr->dir);
9564 free_percpu(tr->last_func_repeats);
9565 free_trace_buffers(tr);
9566 clear_tracing_err_log(tr);
9567
9568 for (i = 0; i < tr->nr_topts; i++) {
9569 kfree(tr->topts[i].topts);
9570 }
9571 kfree(tr->topts);
9572
9573 free_cpumask_var(tr->pipe_cpumask);
9574 free_cpumask_var(tr->tracing_cpumask);
9575 kfree_const(tr->system_names);
9576 kfree(tr->name);
9577 kfree(tr);
9578
9579 return 0;
9580}
9581
9582int trace_array_destroy(struct trace_array *this_tr)
9583{
9584 struct trace_array *tr;
9585
9586 if (!this_tr)
9587 return -EINVAL;
9588
9589 guard(mutex)(&event_mutex);
9590 guard(mutex)(&trace_types_lock);
9591
9592
9593 /* Making sure trace array exists before destroying it. */
9594 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9595 if (tr == this_tr)
9596 return __remove_instance(tr);
9597 }
9598
9599 return -ENODEV;
9600}
9601EXPORT_SYMBOL_GPL(trace_array_destroy);
9602
9603static int instance_rmdir(const char *name)
9604{
9605 struct trace_array *tr;
9606
9607 guard(mutex)(&event_mutex);
9608 guard(mutex)(&trace_types_lock);
9609
9610 tr = trace_array_find(name);
9611 if (!tr)
9612 return -ENODEV;
9613
9614 return __remove_instance(tr);
9615}
9616
9617static __init void create_trace_instances(struct dentry *d_tracer)
9618{
9619 struct trace_array *tr;
9620
9621 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9622 instance_mkdir,
9623 instance_rmdir);
9624 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9625 return;
9626
9627 guard(mutex)(&event_mutex);
9628 guard(mutex)(&trace_types_lock);
9629
9630 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9631 if (!tr->name)
9632 continue;
9633 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9634 "Failed to create instance directory\n"))
9635 return;
9636 }
9637}
9638
9639static void
9640init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9641{
9642 int cpu;
9643
9644 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9645 tr, &show_traces_fops);
9646
9647 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9648 tr, &set_tracer_fops);
9649
9650 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9651 tr, &tracing_cpumask_fops);
9652
9653 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9654 tr, &tracing_iter_fops);
9655
9656 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9657 tr, &tracing_fops);
9658
9659 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9660 tr, &tracing_pipe_fops);
9661
9662 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9663 tr, &tracing_entries_fops);
9664
9665 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9666 tr, &tracing_total_entries_fops);
9667
9668 trace_create_file("free_buffer", 0200, d_tracer,
9669 tr, &tracing_free_buffer_fops);
9670
9671 trace_create_file("trace_marker", 0220, d_tracer,
9672 tr, &tracing_mark_fops);
9673
9674 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9675
9676 trace_create_file("trace_marker_raw", 0220, d_tracer,
9677 tr, &tracing_mark_raw_fops);
9678
9679 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9680 &trace_clock_fops);
9681
9682 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9683 tr, &rb_simple_fops);
9684
9685 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9686 &trace_time_stamp_mode_fops);
9687
9688 tr->buffer_percent = 50;
9689
9690 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9691 tr, &buffer_percent_fops);
9692
9693 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9694 tr, &buffer_subbuf_size_fops);
9695
9696 create_trace_options_dir(tr);
9697
9698#ifdef CONFIG_TRACER_MAX_TRACE
9699 trace_create_maxlat_file(tr, d_tracer);
9700#endif
9701
9702 if (ftrace_create_function_files(tr, d_tracer))
9703 MEM_FAIL(1, "Could not allocate function filter files");
9704
9705 if (tr->range_addr_start) {
9706 trace_create_file("last_boot_info", TRACE_MODE_READ, d_tracer,
9707 tr, &last_boot_fops);
9708#ifdef CONFIG_TRACER_SNAPSHOT
9709 } else {
9710 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9711 tr, &snapshot_fops);
9712#endif
9713 }
9714
9715 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9716 tr, &tracing_err_log_fops);
9717
9718 for_each_tracing_cpu(cpu)
9719 tracing_init_tracefs_percpu(tr, cpu);
9720
9721 ftrace_init_tracefs(tr, d_tracer);
9722}
9723
9724static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9725{
9726 struct vfsmount *mnt;
9727 struct file_system_type *type;
9728
9729 /*
9730 * To maintain backward compatibility for tools that mount
9731 * debugfs to get to the tracing facility, tracefs is automatically
9732 * mounted to the debugfs/tracing directory.
9733 */
9734 type = get_fs_type("tracefs");
9735 if (!type)
9736 return NULL;
9737 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9738 put_filesystem(type);
9739 if (IS_ERR(mnt))
9740 return NULL;
9741 mntget(mnt);
9742
9743 return mnt;
9744}
9745
9746/**
9747 * tracing_init_dentry - initialize top level trace array
9748 *
9749 * This is called when creating files or directories in the tracing
9750 * directory. It is called via fs_initcall() by any of the boot up code
9751 * and expects to return the dentry of the top level tracing directory.
9752 */
9753int tracing_init_dentry(void)
9754{
9755 struct trace_array *tr = &global_trace;
9756
9757 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9758 pr_warn("Tracing disabled due to lockdown\n");
9759 return -EPERM;
9760 }
9761
9762 /* The top level trace array uses NULL as parent */
9763 if (tr->dir)
9764 return 0;
9765
9766 if (WARN_ON(!tracefs_initialized()))
9767 return -ENODEV;
9768
9769 /*
9770 * As there may still be users that expect the tracing
9771 * files to exist in debugfs/tracing, we must automount
9772 * the tracefs file system there, so older tools still
9773 * work with the newer kernel.
9774 */
9775 tr->dir = debugfs_create_automount("tracing", NULL,
9776 trace_automount, NULL);
9777
9778 return 0;
9779}
9780
9781extern struct trace_eval_map *__start_ftrace_eval_maps[];
9782extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9783
9784static struct workqueue_struct *eval_map_wq __initdata;
9785static struct work_struct eval_map_work __initdata;
9786static struct work_struct tracerfs_init_work __initdata;
9787
9788static void __init eval_map_work_func(struct work_struct *work)
9789{
9790 int len;
9791
9792 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9793 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9794}
9795
9796static int __init trace_eval_init(void)
9797{
9798 INIT_WORK(&eval_map_work, eval_map_work_func);
9799
9800 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9801 if (!eval_map_wq) {
9802 pr_err("Unable to allocate eval_map_wq\n");
9803 /* Do work here */
9804 eval_map_work_func(&eval_map_work);
9805 return -ENOMEM;
9806 }
9807
9808 queue_work(eval_map_wq, &eval_map_work);
9809 return 0;
9810}
9811
9812subsys_initcall(trace_eval_init);
9813
9814static int __init trace_eval_sync(void)
9815{
9816 /* Make sure the eval map updates are finished */
9817 if (eval_map_wq)
9818 destroy_workqueue(eval_map_wq);
9819 return 0;
9820}
9821
9822late_initcall_sync(trace_eval_sync);
9823
9824
9825#ifdef CONFIG_MODULES
9826static void trace_module_add_evals(struct module *mod)
9827{
9828 if (!mod->num_trace_evals)
9829 return;
9830
9831 /*
9832 * Modules with bad taint do not have events created, do
9833 * not bother with enums either.
9834 */
9835 if (trace_module_has_bad_taint(mod))
9836 return;
9837
9838 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9839}
9840
9841#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9842static void trace_module_remove_evals(struct module *mod)
9843{
9844 union trace_eval_map_item *map;
9845 union trace_eval_map_item **last = &trace_eval_maps;
9846
9847 if (!mod->num_trace_evals)
9848 return;
9849
9850 guard(mutex)(&trace_eval_mutex);
9851
9852 map = trace_eval_maps;
9853
9854 while (map) {
9855 if (map->head.mod == mod)
9856 break;
9857 map = trace_eval_jmp_to_tail(map);
9858 last = &map->tail.next;
9859 map = map->tail.next;
9860 }
9861 if (!map)
9862 return;
9863
9864 *last = trace_eval_jmp_to_tail(map)->tail.next;
9865 kfree(map);
9866}
9867#else
9868static inline void trace_module_remove_evals(struct module *mod) { }
9869#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9870
9871static int trace_module_notify(struct notifier_block *self,
9872 unsigned long val, void *data)
9873{
9874 struct module *mod = data;
9875
9876 switch (val) {
9877 case MODULE_STATE_COMING:
9878 trace_module_add_evals(mod);
9879 break;
9880 case MODULE_STATE_GOING:
9881 trace_module_remove_evals(mod);
9882 break;
9883 }
9884
9885 return NOTIFY_OK;
9886}
9887
9888static struct notifier_block trace_module_nb = {
9889 .notifier_call = trace_module_notify,
9890 .priority = 0,
9891};
9892#endif /* CONFIG_MODULES */
9893
9894static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9895{
9896
9897 event_trace_init();
9898
9899 init_tracer_tracefs(&global_trace, NULL);
9900 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9901
9902 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9903 &global_trace, &tracing_thresh_fops);
9904
9905 trace_create_file("README", TRACE_MODE_READ, NULL,
9906 NULL, &tracing_readme_fops);
9907
9908 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9909 NULL, &tracing_saved_cmdlines_fops);
9910
9911 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9912 NULL, &tracing_saved_cmdlines_size_fops);
9913
9914 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9915 NULL, &tracing_saved_tgids_fops);
9916
9917 trace_create_eval_file(NULL);
9918
9919#ifdef CONFIG_MODULES
9920 register_module_notifier(&trace_module_nb);
9921#endif
9922
9923#ifdef CONFIG_DYNAMIC_FTRACE
9924 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9925 NULL, &tracing_dyn_info_fops);
9926#endif
9927
9928 create_trace_instances(NULL);
9929
9930 update_tracer_options(&global_trace);
9931}
9932
9933static __init int tracer_init_tracefs(void)
9934{
9935 int ret;
9936
9937 trace_access_lock_init();
9938
9939 ret = tracing_init_dentry();
9940 if (ret)
9941 return 0;
9942
9943 if (eval_map_wq) {
9944 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9945 queue_work(eval_map_wq, &tracerfs_init_work);
9946 } else {
9947 tracer_init_tracefs_work_func(NULL);
9948 }
9949
9950 rv_init_interface();
9951
9952 return 0;
9953}
9954
9955fs_initcall(tracer_init_tracefs);
9956
9957static int trace_die_panic_handler(struct notifier_block *self,
9958 unsigned long ev, void *unused);
9959
9960static struct notifier_block trace_panic_notifier = {
9961 .notifier_call = trace_die_panic_handler,
9962 .priority = INT_MAX - 1,
9963};
9964
9965static struct notifier_block trace_die_notifier = {
9966 .notifier_call = trace_die_panic_handler,
9967 .priority = INT_MAX - 1,
9968};
9969
9970/*
9971 * The idea is to execute the following die/panic callback early, in order
9972 * to avoid showing irrelevant information in the trace (like other panic
9973 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9974 * warnings get disabled (to prevent potential log flooding).
9975 */
9976static int trace_die_panic_handler(struct notifier_block *self,
9977 unsigned long ev, void *unused)
9978{
9979 if (!ftrace_dump_on_oops_enabled())
9980 return NOTIFY_DONE;
9981
9982 /* The die notifier requires DIE_OOPS to trigger */
9983 if (self == &trace_die_notifier && ev != DIE_OOPS)
9984 return NOTIFY_DONE;
9985
9986 ftrace_dump(DUMP_PARAM);
9987
9988 return NOTIFY_DONE;
9989}
9990
9991/*
9992 * printk is set to max of 1024, we really don't need it that big.
9993 * Nothing should be printing 1000 characters anyway.
9994 */
9995#define TRACE_MAX_PRINT 1000
9996
9997/*
9998 * Define here KERN_TRACE so that we have one place to modify
9999 * it if we decide to change what log level the ftrace dump
10000 * should be at.
10001 */
10002#define KERN_TRACE KERN_EMERG
10003
10004void
10005trace_printk_seq(struct trace_seq *s)
10006{
10007 /* Probably should print a warning here. */
10008 if (s->seq.len >= TRACE_MAX_PRINT)
10009 s->seq.len = TRACE_MAX_PRINT;
10010
10011 /*
10012 * More paranoid code. Although the buffer size is set to
10013 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10014 * an extra layer of protection.
10015 */
10016 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10017 s->seq.len = s->seq.size - 1;
10018
10019 /* should be zero ended, but we are paranoid. */
10020 s->buffer[s->seq.len] = 0;
10021
10022 printk(KERN_TRACE "%s", s->buffer);
10023
10024 trace_seq_init(s);
10025}
10026
10027static void trace_init_iter(struct trace_iterator *iter, struct trace_array *tr)
10028{
10029 iter->tr = tr;
10030 iter->trace = iter->tr->current_trace;
10031 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10032 iter->array_buffer = &tr->array_buffer;
10033
10034 if (iter->trace && iter->trace->open)
10035 iter->trace->open(iter);
10036
10037 /* Annotate start of buffers if we had overruns */
10038 if (ring_buffer_overruns(iter->array_buffer->buffer))
10039 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10040
10041 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10042 if (trace_clocks[iter->tr->clock_id].in_ns)
10043 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10044
10045 /* Can not use kmalloc for iter.temp and iter.fmt */
10046 iter->temp = static_temp_buf;
10047 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10048 iter->fmt = static_fmt_buf;
10049 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10050}
10051
10052void trace_init_global_iter(struct trace_iterator *iter)
10053{
10054 trace_init_iter(iter, &global_trace);
10055}
10056
10057static void ftrace_dump_one(struct trace_array *tr, enum ftrace_dump_mode dump_mode)
10058{
10059 /* use static because iter can be a bit big for the stack */
10060 static struct trace_iterator iter;
10061 unsigned int old_userobj;
10062 unsigned long flags;
10063 int cnt = 0, cpu;
10064
10065 /*
10066 * Always turn off tracing when we dump.
10067 * We don't need to show trace output of what happens
10068 * between multiple crashes.
10069 *
10070 * If the user does a sysrq-z, then they can re-enable
10071 * tracing with echo 1 > tracing_on.
10072 */
10073 tracer_tracing_off(tr);
10074
10075 local_irq_save(flags);
10076
10077 /* Simulate the iterator */
10078 trace_init_iter(&iter, tr);
10079
10080 for_each_tracing_cpu(cpu) {
10081 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10082 }
10083
10084 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10085
10086 /* don't look at user memory in panic mode */
10087 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10088
10089 if (dump_mode == DUMP_ORIG)
10090 iter.cpu_file = raw_smp_processor_id();
10091 else
10092 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10093
10094 if (tr == &global_trace)
10095 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10096 else
10097 printk(KERN_TRACE "Dumping ftrace instance %s buffer:\n", tr->name);
10098
10099 /* Did function tracer already get disabled? */
10100 if (ftrace_is_dead()) {
10101 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10102 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10103 }
10104
10105 /*
10106 * We need to stop all tracing on all CPUS to read
10107 * the next buffer. This is a bit expensive, but is
10108 * not done often. We fill all what we can read,
10109 * and then release the locks again.
10110 */
10111
10112 while (!trace_empty(&iter)) {
10113
10114 if (!cnt)
10115 printk(KERN_TRACE "---------------------------------\n");
10116
10117 cnt++;
10118
10119 trace_iterator_reset(&iter);
10120 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10121
10122 if (trace_find_next_entry_inc(&iter) != NULL) {
10123 int ret;
10124
10125 ret = print_trace_line(&iter);
10126 if (ret != TRACE_TYPE_NO_CONSUME)
10127 trace_consume(&iter);
10128 }
10129 touch_nmi_watchdog();
10130
10131 trace_printk_seq(&iter.seq);
10132 }
10133
10134 if (!cnt)
10135 printk(KERN_TRACE " (ftrace buffer empty)\n");
10136 else
10137 printk(KERN_TRACE "---------------------------------\n");
10138
10139 tr->trace_flags |= old_userobj;
10140
10141 for_each_tracing_cpu(cpu) {
10142 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10143 }
10144 local_irq_restore(flags);
10145}
10146
10147static void ftrace_dump_by_param(void)
10148{
10149 bool first_param = true;
10150 char dump_param[MAX_TRACER_SIZE];
10151 char *buf, *token, *inst_name;
10152 struct trace_array *tr;
10153
10154 strscpy(dump_param, ftrace_dump_on_oops, MAX_TRACER_SIZE);
10155 buf = dump_param;
10156
10157 while ((token = strsep(&buf, ",")) != NULL) {
10158 if (first_param) {
10159 first_param = false;
10160 if (!strcmp("0", token))
10161 continue;
10162 else if (!strcmp("1", token)) {
10163 ftrace_dump_one(&global_trace, DUMP_ALL);
10164 continue;
10165 }
10166 else if (!strcmp("2", token) ||
10167 !strcmp("orig_cpu", token)) {
10168 ftrace_dump_one(&global_trace, DUMP_ORIG);
10169 continue;
10170 }
10171 }
10172
10173 inst_name = strsep(&token, "=");
10174 tr = trace_array_find(inst_name);
10175 if (!tr) {
10176 printk(KERN_TRACE "Instance %s not found\n", inst_name);
10177 continue;
10178 }
10179
10180 if (token && (!strcmp("2", token) ||
10181 !strcmp("orig_cpu", token)))
10182 ftrace_dump_one(tr, DUMP_ORIG);
10183 else
10184 ftrace_dump_one(tr, DUMP_ALL);
10185 }
10186}
10187
10188void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10189{
10190 static atomic_t dump_running;
10191
10192 /* Only allow one dump user at a time. */
10193 if (atomic_inc_return(&dump_running) != 1) {
10194 atomic_dec(&dump_running);
10195 return;
10196 }
10197
10198 switch (oops_dump_mode) {
10199 case DUMP_ALL:
10200 ftrace_dump_one(&global_trace, DUMP_ALL);
10201 break;
10202 case DUMP_ORIG:
10203 ftrace_dump_one(&global_trace, DUMP_ORIG);
10204 break;
10205 case DUMP_PARAM:
10206 ftrace_dump_by_param();
10207 break;
10208 case DUMP_NONE:
10209 break;
10210 default:
10211 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10212 ftrace_dump_one(&global_trace, DUMP_ALL);
10213 }
10214
10215 atomic_dec(&dump_running);
10216}
10217EXPORT_SYMBOL_GPL(ftrace_dump);
10218
10219#define WRITE_BUFSIZE 4096
10220
10221ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10222 size_t count, loff_t *ppos,
10223 int (*createfn)(const char *))
10224{
10225 char *kbuf, *buf, *tmp;
10226 int ret = 0;
10227 size_t done = 0;
10228 size_t size;
10229
10230 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10231 if (!kbuf)
10232 return -ENOMEM;
10233
10234 while (done < count) {
10235 size = count - done;
10236
10237 if (size >= WRITE_BUFSIZE)
10238 size = WRITE_BUFSIZE - 1;
10239
10240 if (copy_from_user(kbuf, buffer + done, size)) {
10241 ret = -EFAULT;
10242 goto out;
10243 }
10244 kbuf[size] = '\0';
10245 buf = kbuf;
10246 do {
10247 tmp = strchr(buf, '\n');
10248 if (tmp) {
10249 *tmp = '\0';
10250 size = tmp - buf + 1;
10251 } else {
10252 size = strlen(buf);
10253 if (done + size < count) {
10254 if (buf != kbuf)
10255 break;
10256 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10257 pr_warn("Line length is too long: Should be less than %d\n",
10258 WRITE_BUFSIZE - 2);
10259 ret = -EINVAL;
10260 goto out;
10261 }
10262 }
10263 done += size;
10264
10265 /* Remove comments */
10266 tmp = strchr(buf, '#');
10267
10268 if (tmp)
10269 *tmp = '\0';
10270
10271 ret = createfn(buf);
10272 if (ret)
10273 goto out;
10274 buf += size;
10275
10276 } while (done < count);
10277 }
10278 ret = done;
10279
10280out:
10281 kfree(kbuf);
10282
10283 return ret;
10284}
10285
10286#ifdef CONFIG_TRACER_MAX_TRACE
10287__init static bool tr_needs_alloc_snapshot(const char *name)
10288{
10289 char *test;
10290 int len = strlen(name);
10291 bool ret;
10292
10293 if (!boot_snapshot_index)
10294 return false;
10295
10296 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10297 boot_snapshot_info[len] == '\t')
10298 return true;
10299
10300 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10301 if (!test)
10302 return false;
10303
10304 sprintf(test, "\t%s\t", name);
10305 ret = strstr(boot_snapshot_info, test) == NULL;
10306 kfree(test);
10307 return ret;
10308}
10309
10310__init static void do_allocate_snapshot(const char *name)
10311{
10312 if (!tr_needs_alloc_snapshot(name))
10313 return;
10314
10315 /*
10316 * When allocate_snapshot is set, the next call to
10317 * allocate_trace_buffers() (called by trace_array_get_by_name())
10318 * will allocate the snapshot buffer. That will alse clear
10319 * this flag.
10320 */
10321 allocate_snapshot = true;
10322}
10323#else
10324static inline void do_allocate_snapshot(const char *name) { }
10325#endif
10326
10327__init static void enable_instances(void)
10328{
10329 struct trace_array *tr;
10330 char *curr_str;
10331 char *name;
10332 char *str;
10333 char *tok;
10334
10335 /* A tab is always appended */
10336 boot_instance_info[boot_instance_index - 1] = '\0';
10337 str = boot_instance_info;
10338
10339 while ((curr_str = strsep(&str, "\t"))) {
10340 phys_addr_t start = 0;
10341 phys_addr_t size = 0;
10342 unsigned long addr = 0;
10343 bool traceprintk = false;
10344 bool traceoff = false;
10345 char *flag_delim;
10346 char *addr_delim;
10347
10348 tok = strsep(&curr_str, ",");
10349
10350 flag_delim = strchr(tok, '^');
10351 addr_delim = strchr(tok, '@');
10352
10353 if (addr_delim)
10354 *addr_delim++ = '\0';
10355
10356 if (flag_delim)
10357 *flag_delim++ = '\0';
10358
10359 name = tok;
10360
10361 if (flag_delim) {
10362 char *flag;
10363
10364 while ((flag = strsep(&flag_delim, "^"))) {
10365 if (strcmp(flag, "traceoff") == 0) {
10366 traceoff = true;
10367 } else if ((strcmp(flag, "printk") == 0) ||
10368 (strcmp(flag, "traceprintk") == 0) ||
10369 (strcmp(flag, "trace_printk") == 0)) {
10370 traceprintk = true;
10371 } else {
10372 pr_info("Tracing: Invalid instance flag '%s' for %s\n",
10373 flag, name);
10374 }
10375 }
10376 }
10377
10378 tok = addr_delim;
10379 if (tok && isdigit(*tok)) {
10380 start = memparse(tok, &tok);
10381 if (!start) {
10382 pr_warn("Tracing: Invalid boot instance address for %s\n",
10383 name);
10384 continue;
10385 }
10386 if (*tok != ':') {
10387 pr_warn("Tracing: No size specified for instance %s\n", name);
10388 continue;
10389 }
10390 tok++;
10391 size = memparse(tok, &tok);
10392 if (!size) {
10393 pr_warn("Tracing: Invalid boot instance size for %s\n",
10394 name);
10395 continue;
10396 }
10397 } else if (tok) {
10398 if (!reserve_mem_find_by_name(tok, &start, &size)) {
10399 start = 0;
10400 pr_warn("Failed to map boot instance %s to %s\n", name, tok);
10401 continue;
10402 }
10403 }
10404
10405 if (start) {
10406 addr = map_pages(start, size);
10407 if (addr) {
10408 pr_info("Tracing: mapped boot instance %s at physical memory %pa of size 0x%lx\n",
10409 name, &start, (unsigned long)size);
10410 } else {
10411 pr_warn("Tracing: Failed to map boot instance %s\n", name);
10412 continue;
10413 }
10414 } else {
10415 /* Only non mapped buffers have snapshot buffers */
10416 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10417 do_allocate_snapshot(name);
10418 }
10419
10420 tr = trace_array_create_systems(name, NULL, addr, size);
10421 if (IS_ERR(tr)) {
10422 pr_warn("Tracing: Failed to create instance buffer %s\n", curr_str);
10423 continue;
10424 }
10425
10426 if (traceoff)
10427 tracer_tracing_off(tr);
10428
10429 if (traceprintk)
10430 update_printk_trace(tr);
10431
10432 /*
10433 * If start is set, then this is a mapped buffer, and
10434 * cannot be deleted by user space, so keep the reference
10435 * to it.
10436 */
10437 if (start) {
10438 tr->flags |= TRACE_ARRAY_FL_BOOT;
10439 tr->ref++;
10440 }
10441
10442 while ((tok = strsep(&curr_str, ","))) {
10443 early_enable_events(tr, tok, true);
10444 }
10445 }
10446}
10447
10448__init static int tracer_alloc_buffers(void)
10449{
10450 int ring_buf_size;
10451 int ret = -ENOMEM;
10452
10453
10454 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10455 pr_warn("Tracing disabled due to lockdown\n");
10456 return -EPERM;
10457 }
10458
10459 /*
10460 * Make sure we don't accidentally add more trace options
10461 * than we have bits for.
10462 */
10463 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10464
10465 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10466 goto out;
10467
10468 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10469 goto out_free_buffer_mask;
10470
10471 /* Only allocate trace_printk buffers if a trace_printk exists */
10472 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10473 /* Must be called before global_trace.buffer is allocated */
10474 trace_printk_init_buffers();
10475
10476 /* To save memory, keep the ring buffer size to its minimum */
10477 if (global_trace.ring_buffer_expanded)
10478 ring_buf_size = trace_buf_size;
10479 else
10480 ring_buf_size = 1;
10481
10482 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10483 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10484
10485 raw_spin_lock_init(&global_trace.start_lock);
10486
10487 /*
10488 * The prepare callbacks allocates some memory for the ring buffer. We
10489 * don't free the buffer if the CPU goes down. If we were to free
10490 * the buffer, then the user would lose any trace that was in the
10491 * buffer. The memory will be removed once the "instance" is removed.
10492 */
10493 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10494 "trace/RB:prepare", trace_rb_cpu_prepare,
10495 NULL);
10496 if (ret < 0)
10497 goto out_free_cpumask;
10498 /* Used for event triggers */
10499 ret = -ENOMEM;
10500 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10501 if (!temp_buffer)
10502 goto out_rm_hp_state;
10503
10504 if (trace_create_savedcmd() < 0)
10505 goto out_free_temp_buffer;
10506
10507 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10508 goto out_free_savedcmd;
10509
10510 /* TODO: make the number of buffers hot pluggable with CPUS */
10511 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10512 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10513 goto out_free_pipe_cpumask;
10514 }
10515 if (global_trace.buffer_disabled)
10516 tracing_off();
10517
10518 if (trace_boot_clock) {
10519 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10520 if (ret < 0)
10521 pr_warn("Trace clock %s not defined, going back to default\n",
10522 trace_boot_clock);
10523 }
10524
10525 /*
10526 * register_tracer() might reference current_trace, so it
10527 * needs to be set before we register anything. This is
10528 * just a bootstrap of current_trace anyway.
10529 */
10530 global_trace.current_trace = &nop_trace;
10531
10532 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10533#ifdef CONFIG_TRACER_MAX_TRACE
10534 spin_lock_init(&global_trace.snapshot_trigger_lock);
10535#endif
10536 ftrace_init_global_array_ops(&global_trace);
10537
10538 init_trace_flags_index(&global_trace);
10539
10540 register_tracer(&nop_trace);
10541
10542 /* Function tracing may start here (via kernel command line) */
10543 init_function_trace();
10544
10545 /* All seems OK, enable tracing */
10546 tracing_disabled = 0;
10547
10548 atomic_notifier_chain_register(&panic_notifier_list,
10549 &trace_panic_notifier);
10550
10551 register_die_notifier(&trace_die_notifier);
10552
10553 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10554
10555 INIT_LIST_HEAD(&global_trace.systems);
10556 INIT_LIST_HEAD(&global_trace.events);
10557 INIT_LIST_HEAD(&global_trace.hist_vars);
10558 INIT_LIST_HEAD(&global_trace.err_log);
10559 list_add(&global_trace.list, &ftrace_trace_arrays);
10560
10561 apply_trace_boot_options();
10562
10563 register_snapshot_cmd();
10564
10565 return 0;
10566
10567out_free_pipe_cpumask:
10568 free_cpumask_var(global_trace.pipe_cpumask);
10569out_free_savedcmd:
10570 trace_free_saved_cmdlines_buffer();
10571out_free_temp_buffer:
10572 ring_buffer_free(temp_buffer);
10573out_rm_hp_state:
10574 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10575out_free_cpumask:
10576 free_cpumask_var(global_trace.tracing_cpumask);
10577out_free_buffer_mask:
10578 free_cpumask_var(tracing_buffer_mask);
10579out:
10580 return ret;
10581}
10582
10583void __init ftrace_boot_snapshot(void)
10584{
10585#ifdef CONFIG_TRACER_MAX_TRACE
10586 struct trace_array *tr;
10587
10588 if (!snapshot_at_boot)
10589 return;
10590
10591 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10592 if (!tr->allocated_snapshot)
10593 continue;
10594
10595 tracing_snapshot_instance(tr);
10596 trace_array_puts(tr, "** Boot snapshot taken **\n");
10597 }
10598#endif
10599}
10600
10601void __init early_trace_init(void)
10602{
10603 if (tracepoint_printk) {
10604 tracepoint_print_iter =
10605 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10606 if (MEM_FAIL(!tracepoint_print_iter,
10607 "Failed to allocate trace iterator\n"))
10608 tracepoint_printk = 0;
10609 else
10610 static_key_enable(&tracepoint_printk_key.key);
10611 }
10612 tracer_alloc_buffers();
10613
10614 init_events();
10615}
10616
10617void __init trace_init(void)
10618{
10619 trace_event_init();
10620
10621 if (boot_instance_index)
10622 enable_instances();
10623}
10624
10625__init static void clear_boot_tracer(void)
10626{
10627 /*
10628 * The default tracer at boot buffer is an init section.
10629 * This function is called in lateinit. If we did not
10630 * find the boot tracer, then clear it out, to prevent
10631 * later registration from accessing the buffer that is
10632 * about to be freed.
10633 */
10634 if (!default_bootup_tracer)
10635 return;
10636
10637 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10638 default_bootup_tracer);
10639 default_bootup_tracer = NULL;
10640}
10641
10642#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10643__init static void tracing_set_default_clock(void)
10644{
10645 /* sched_clock_stable() is determined in late_initcall */
10646 if (!trace_boot_clock && !sched_clock_stable()) {
10647 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10648 pr_warn("Can not set tracing clock due to lockdown\n");
10649 return;
10650 }
10651
10652 printk(KERN_WARNING
10653 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10654 "If you want to keep using the local clock, then add:\n"
10655 " \"trace_clock=local\"\n"
10656 "on the kernel command line\n");
10657 tracing_set_clock(&global_trace, "global");
10658 }
10659}
10660#else
10661static inline void tracing_set_default_clock(void) { }
10662#endif
10663
10664__init static int late_trace_init(void)
10665{
10666 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10667 static_key_disable(&tracepoint_printk_key.key);
10668 tracepoint_printk = 0;
10669 }
10670
10671 tracing_set_default_clock();
10672 clear_boot_tracer();
10673 return 0;
10674}
10675
10676late_initcall_sync(late_trace_init);
1/*
2 * ring buffer based function tracer
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
13 */
14#include <linux/ring_buffer.h>
15#include <generated/utsrelease.h>
16#include <linux/stacktrace.h>
17#include <linux/writeback.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/notifier.h>
21#include <linux/irqflags.h>
22#include <linux/debugfs.h>
23#include <linux/pagemap.h>
24#include <linux/hardirq.h>
25#include <linux/linkage.h>
26#include <linux/uaccess.h>
27#include <linux/kprobes.h>
28#include <linux/ftrace.h>
29#include <linux/module.h>
30#include <linux/percpu.h>
31#include <linux/splice.h>
32#include <linux/kdebug.h>
33#include <linux/string.h>
34#include <linux/rwsem.h>
35#include <linux/slab.h>
36#include <linux/ctype.h>
37#include <linux/init.h>
38#include <linux/poll.h>
39#include <linux/nmi.h>
40#include <linux/fs.h>
41
42#include "trace.h"
43#include "trace_output.h"
44
45/*
46 * On boot up, the ring buffer is set to the minimum size, so that
47 * we do not waste memory on systems that are not using tracing.
48 */
49int ring_buffer_expanded;
50
51/*
52 * We need to change this state when a selftest is running.
53 * A selftest will lurk into the ring-buffer to count the
54 * entries inserted during the selftest although some concurrent
55 * insertions into the ring-buffer such as trace_printk could occurred
56 * at the same time, giving false positive or negative results.
57 */
58static bool __read_mostly tracing_selftest_running;
59
60/*
61 * If a tracer is running, we do not want to run SELFTEST.
62 */
63bool __read_mostly tracing_selftest_disabled;
64
65/* For tracers that don't implement custom flags */
66static struct tracer_opt dummy_tracer_opt[] = {
67 { }
68};
69
70static struct tracer_flags dummy_tracer_flags = {
71 .val = 0,
72 .opts = dummy_tracer_opt
73};
74
75static int dummy_set_flag(u32 old_flags, u32 bit, int set)
76{
77 return 0;
78}
79
80/*
81 * Kill all tracing for good (never come back).
82 * It is initialized to 1 but will turn to zero if the initialization
83 * of the tracer is successful. But that is the only place that sets
84 * this back to zero.
85 */
86static int tracing_disabled = 1;
87
88DEFINE_PER_CPU(int, ftrace_cpu_disabled);
89
90cpumask_var_t __read_mostly tracing_buffer_mask;
91
92/*
93 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
94 *
95 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
96 * is set, then ftrace_dump is called. This will output the contents
97 * of the ftrace buffers to the console. This is very useful for
98 * capturing traces that lead to crashes and outputing it to a
99 * serial console.
100 *
101 * It is default off, but you can enable it with either specifying
102 * "ftrace_dump_on_oops" in the kernel command line, or setting
103 * /proc/sys/kernel/ftrace_dump_on_oops
104 * Set 1 if you want to dump buffers of all CPUs
105 * Set 2 if you want to dump the buffer of the CPU that triggered oops
106 */
107
108enum ftrace_dump_mode ftrace_dump_on_oops;
109
110static int tracing_set_tracer(const char *buf);
111
112#define MAX_TRACER_SIZE 100
113static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
114static char *default_bootup_tracer;
115
116static int __init set_cmdline_ftrace(char *str)
117{
118 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
119 default_bootup_tracer = bootup_tracer_buf;
120 /* We are using ftrace early, expand it */
121 ring_buffer_expanded = 1;
122 return 1;
123}
124__setup("ftrace=", set_cmdline_ftrace);
125
126static int __init set_ftrace_dump_on_oops(char *str)
127{
128 if (*str++ != '=' || !*str) {
129 ftrace_dump_on_oops = DUMP_ALL;
130 return 1;
131 }
132
133 if (!strcmp("orig_cpu", str)) {
134 ftrace_dump_on_oops = DUMP_ORIG;
135 return 1;
136 }
137
138 return 0;
139}
140__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
141
142unsigned long long ns2usecs(cycle_t nsec)
143{
144 nsec += 500;
145 do_div(nsec, 1000);
146 return nsec;
147}
148
149/*
150 * The global_trace is the descriptor that holds the tracing
151 * buffers for the live tracing. For each CPU, it contains
152 * a link list of pages that will store trace entries. The
153 * page descriptor of the pages in the memory is used to hold
154 * the link list by linking the lru item in the page descriptor
155 * to each of the pages in the buffer per CPU.
156 *
157 * For each active CPU there is a data field that holds the
158 * pages for the buffer for that CPU. Each CPU has the same number
159 * of pages allocated for its buffer.
160 */
161static struct trace_array global_trace;
162
163static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
164
165int filter_current_check_discard(struct ring_buffer *buffer,
166 struct ftrace_event_call *call, void *rec,
167 struct ring_buffer_event *event)
168{
169 return filter_check_discard(call, rec, buffer, event);
170}
171EXPORT_SYMBOL_GPL(filter_current_check_discard);
172
173cycle_t ftrace_now(int cpu)
174{
175 u64 ts;
176
177 /* Early boot up does not have a buffer yet */
178 if (!global_trace.buffer)
179 return trace_clock_local();
180
181 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
182 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
183
184 return ts;
185}
186
187/*
188 * The max_tr is used to snapshot the global_trace when a maximum
189 * latency is reached. Some tracers will use this to store a maximum
190 * trace while it continues examining live traces.
191 *
192 * The buffers for the max_tr are set up the same as the global_trace.
193 * When a snapshot is taken, the link list of the max_tr is swapped
194 * with the link list of the global_trace and the buffers are reset for
195 * the global_trace so the tracing can continue.
196 */
197static struct trace_array max_tr;
198
199static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
200
201/* tracer_enabled is used to toggle activation of a tracer */
202static int tracer_enabled = 1;
203
204/**
205 * tracing_is_enabled - return tracer_enabled status
206 *
207 * This function is used by other tracers to know the status
208 * of the tracer_enabled flag. Tracers may use this function
209 * to know if it should enable their features when starting
210 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
211 */
212int tracing_is_enabled(void)
213{
214 return tracer_enabled;
215}
216
217/*
218 * trace_buf_size is the size in bytes that is allocated
219 * for a buffer. Note, the number of bytes is always rounded
220 * to page size.
221 *
222 * This number is purposely set to a low number of 16384.
223 * If the dump on oops happens, it will be much appreciated
224 * to not have to wait for all that output. Anyway this can be
225 * boot time and run time configurable.
226 */
227#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
228
229static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
230
231/* trace_types holds a link list of available tracers. */
232static struct tracer *trace_types __read_mostly;
233
234/* current_trace points to the tracer that is currently active */
235static struct tracer *current_trace __read_mostly;
236
237/*
238 * trace_types_lock is used to protect the trace_types list.
239 */
240static DEFINE_MUTEX(trace_types_lock);
241
242/*
243 * serialize the access of the ring buffer
244 *
245 * ring buffer serializes readers, but it is low level protection.
246 * The validity of the events (which returns by ring_buffer_peek() ..etc)
247 * are not protected by ring buffer.
248 *
249 * The content of events may become garbage if we allow other process consumes
250 * these events concurrently:
251 * A) the page of the consumed events may become a normal page
252 * (not reader page) in ring buffer, and this page will be rewrited
253 * by events producer.
254 * B) The page of the consumed events may become a page for splice_read,
255 * and this page will be returned to system.
256 *
257 * These primitives allow multi process access to different cpu ring buffer
258 * concurrently.
259 *
260 * These primitives don't distinguish read-only and read-consume access.
261 * Multi read-only access are also serialized.
262 */
263
264#ifdef CONFIG_SMP
265static DECLARE_RWSEM(all_cpu_access_lock);
266static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
267
268static inline void trace_access_lock(int cpu)
269{
270 if (cpu == TRACE_PIPE_ALL_CPU) {
271 /* gain it for accessing the whole ring buffer. */
272 down_write(&all_cpu_access_lock);
273 } else {
274 /* gain it for accessing a cpu ring buffer. */
275
276 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
277 down_read(&all_cpu_access_lock);
278
279 /* Secondly block other access to this @cpu ring buffer. */
280 mutex_lock(&per_cpu(cpu_access_lock, cpu));
281 }
282}
283
284static inline void trace_access_unlock(int cpu)
285{
286 if (cpu == TRACE_PIPE_ALL_CPU) {
287 up_write(&all_cpu_access_lock);
288 } else {
289 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
290 up_read(&all_cpu_access_lock);
291 }
292}
293
294static inline void trace_access_lock_init(void)
295{
296 int cpu;
297
298 for_each_possible_cpu(cpu)
299 mutex_init(&per_cpu(cpu_access_lock, cpu));
300}
301
302#else
303
304static DEFINE_MUTEX(access_lock);
305
306static inline void trace_access_lock(int cpu)
307{
308 (void)cpu;
309 mutex_lock(&access_lock);
310}
311
312static inline void trace_access_unlock(int cpu)
313{
314 (void)cpu;
315 mutex_unlock(&access_lock);
316}
317
318static inline void trace_access_lock_init(void)
319{
320}
321
322#endif
323
324/* trace_wait is a waitqueue for tasks blocked on trace_poll */
325static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
326
327/* trace_flags holds trace_options default values */
328unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
329 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
330 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
331 TRACE_ITER_IRQ_INFO;
332
333static int trace_stop_count;
334static DEFINE_RAW_SPINLOCK(tracing_start_lock);
335
336static void wakeup_work_handler(struct work_struct *work)
337{
338 wake_up(&trace_wait);
339}
340
341static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
342
343/**
344 * tracing_on - enable tracing buffers
345 *
346 * This function enables tracing buffers that may have been
347 * disabled with tracing_off.
348 */
349void tracing_on(void)
350{
351 if (global_trace.buffer)
352 ring_buffer_record_on(global_trace.buffer);
353 /*
354 * This flag is only looked at when buffers haven't been
355 * allocated yet. We don't really care about the race
356 * between setting this flag and actually turning
357 * on the buffer.
358 */
359 global_trace.buffer_disabled = 0;
360}
361EXPORT_SYMBOL_GPL(tracing_on);
362
363/**
364 * tracing_off - turn off tracing buffers
365 *
366 * This function stops the tracing buffers from recording data.
367 * It does not disable any overhead the tracers themselves may
368 * be causing. This function simply causes all recording to
369 * the ring buffers to fail.
370 */
371void tracing_off(void)
372{
373 if (global_trace.buffer)
374 ring_buffer_record_off(global_trace.buffer);
375 /*
376 * This flag is only looked at when buffers haven't been
377 * allocated yet. We don't really care about the race
378 * between setting this flag and actually turning
379 * on the buffer.
380 */
381 global_trace.buffer_disabled = 1;
382}
383EXPORT_SYMBOL_GPL(tracing_off);
384
385/**
386 * tracing_is_on - show state of ring buffers enabled
387 */
388int tracing_is_on(void)
389{
390 if (global_trace.buffer)
391 return ring_buffer_record_is_on(global_trace.buffer);
392 return !global_trace.buffer_disabled;
393}
394EXPORT_SYMBOL_GPL(tracing_is_on);
395
396/**
397 * trace_wake_up - wake up tasks waiting for trace input
398 *
399 * Schedules a delayed work to wake up any task that is blocked on the
400 * trace_wait queue. These is used with trace_poll for tasks polling the
401 * trace.
402 */
403void trace_wake_up(void)
404{
405 const unsigned long delay = msecs_to_jiffies(2);
406
407 if (trace_flags & TRACE_ITER_BLOCK)
408 return;
409 schedule_delayed_work(&wakeup_work, delay);
410}
411
412static int __init set_buf_size(char *str)
413{
414 unsigned long buf_size;
415
416 if (!str)
417 return 0;
418 buf_size = memparse(str, &str);
419 /* nr_entries can not be zero */
420 if (buf_size == 0)
421 return 0;
422 trace_buf_size = buf_size;
423 return 1;
424}
425__setup("trace_buf_size=", set_buf_size);
426
427static int __init set_tracing_thresh(char *str)
428{
429 unsigned long threshhold;
430 int ret;
431
432 if (!str)
433 return 0;
434 ret = strict_strtoul(str, 0, &threshhold);
435 if (ret < 0)
436 return 0;
437 tracing_thresh = threshhold * 1000;
438 return 1;
439}
440__setup("tracing_thresh=", set_tracing_thresh);
441
442unsigned long nsecs_to_usecs(unsigned long nsecs)
443{
444 return nsecs / 1000;
445}
446
447/* These must match the bit postions in trace_iterator_flags */
448static const char *trace_options[] = {
449 "print-parent",
450 "sym-offset",
451 "sym-addr",
452 "verbose",
453 "raw",
454 "hex",
455 "bin",
456 "block",
457 "stacktrace",
458 "trace_printk",
459 "ftrace_preempt",
460 "branch",
461 "annotate",
462 "userstacktrace",
463 "sym-userobj",
464 "printk-msg-only",
465 "context-info",
466 "latency-format",
467 "sleep-time",
468 "graph-time",
469 "record-cmd",
470 "overwrite",
471 "disable_on_free",
472 "irq-info",
473 NULL
474};
475
476static struct {
477 u64 (*func)(void);
478 const char *name;
479} trace_clocks[] = {
480 { trace_clock_local, "local" },
481 { trace_clock_global, "global" },
482 { trace_clock_counter, "counter" },
483};
484
485int trace_clock_id;
486
487/*
488 * trace_parser_get_init - gets the buffer for trace parser
489 */
490int trace_parser_get_init(struct trace_parser *parser, int size)
491{
492 memset(parser, 0, sizeof(*parser));
493
494 parser->buffer = kmalloc(size, GFP_KERNEL);
495 if (!parser->buffer)
496 return 1;
497
498 parser->size = size;
499 return 0;
500}
501
502/*
503 * trace_parser_put - frees the buffer for trace parser
504 */
505void trace_parser_put(struct trace_parser *parser)
506{
507 kfree(parser->buffer);
508}
509
510/*
511 * trace_get_user - reads the user input string separated by space
512 * (matched by isspace(ch))
513 *
514 * For each string found the 'struct trace_parser' is updated,
515 * and the function returns.
516 *
517 * Returns number of bytes read.
518 *
519 * See kernel/trace/trace.h for 'struct trace_parser' details.
520 */
521int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
522 size_t cnt, loff_t *ppos)
523{
524 char ch;
525 size_t read = 0;
526 ssize_t ret;
527
528 if (!*ppos)
529 trace_parser_clear(parser);
530
531 ret = get_user(ch, ubuf++);
532 if (ret)
533 goto out;
534
535 read++;
536 cnt--;
537
538 /*
539 * The parser is not finished with the last write,
540 * continue reading the user input without skipping spaces.
541 */
542 if (!parser->cont) {
543 /* skip white space */
544 while (cnt && isspace(ch)) {
545 ret = get_user(ch, ubuf++);
546 if (ret)
547 goto out;
548 read++;
549 cnt--;
550 }
551
552 /* only spaces were written */
553 if (isspace(ch)) {
554 *ppos += read;
555 ret = read;
556 goto out;
557 }
558
559 parser->idx = 0;
560 }
561
562 /* read the non-space input */
563 while (cnt && !isspace(ch)) {
564 if (parser->idx < parser->size - 1)
565 parser->buffer[parser->idx++] = ch;
566 else {
567 ret = -EINVAL;
568 goto out;
569 }
570 ret = get_user(ch, ubuf++);
571 if (ret)
572 goto out;
573 read++;
574 cnt--;
575 }
576
577 /* We either got finished input or we have to wait for another call. */
578 if (isspace(ch)) {
579 parser->buffer[parser->idx] = 0;
580 parser->cont = false;
581 } else {
582 parser->cont = true;
583 parser->buffer[parser->idx++] = ch;
584 }
585
586 *ppos += read;
587 ret = read;
588
589out:
590 return ret;
591}
592
593ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
594{
595 int len;
596 int ret;
597
598 if (!cnt)
599 return 0;
600
601 if (s->len <= s->readpos)
602 return -EBUSY;
603
604 len = s->len - s->readpos;
605 if (cnt > len)
606 cnt = len;
607 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
608 if (ret == cnt)
609 return -EFAULT;
610
611 cnt -= ret;
612
613 s->readpos += cnt;
614 return cnt;
615}
616
617static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
618{
619 int len;
620
621 if (s->len <= s->readpos)
622 return -EBUSY;
623
624 len = s->len - s->readpos;
625 if (cnt > len)
626 cnt = len;
627 memcpy(buf, s->buffer + s->readpos, cnt);
628
629 s->readpos += cnt;
630 return cnt;
631}
632
633/*
634 * ftrace_max_lock is used to protect the swapping of buffers
635 * when taking a max snapshot. The buffers themselves are
636 * protected by per_cpu spinlocks. But the action of the swap
637 * needs its own lock.
638 *
639 * This is defined as a arch_spinlock_t in order to help
640 * with performance when lockdep debugging is enabled.
641 *
642 * It is also used in other places outside the update_max_tr
643 * so it needs to be defined outside of the
644 * CONFIG_TRACER_MAX_TRACE.
645 */
646static arch_spinlock_t ftrace_max_lock =
647 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
648
649unsigned long __read_mostly tracing_thresh;
650
651#ifdef CONFIG_TRACER_MAX_TRACE
652unsigned long __read_mostly tracing_max_latency;
653
654/*
655 * Copy the new maximum trace into the separate maximum-trace
656 * structure. (this way the maximum trace is permanently saved,
657 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
658 */
659static void
660__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
661{
662 struct trace_array_cpu *data = tr->data[cpu];
663 struct trace_array_cpu *max_data;
664
665 max_tr.cpu = cpu;
666 max_tr.time_start = data->preempt_timestamp;
667
668 max_data = max_tr.data[cpu];
669 max_data->saved_latency = tracing_max_latency;
670 max_data->critical_start = data->critical_start;
671 max_data->critical_end = data->critical_end;
672
673 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
674 max_data->pid = tsk->pid;
675 max_data->uid = task_uid(tsk);
676 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
677 max_data->policy = tsk->policy;
678 max_data->rt_priority = tsk->rt_priority;
679
680 /* record this tasks comm */
681 tracing_record_cmdline(tsk);
682}
683
684/**
685 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
686 * @tr: tracer
687 * @tsk: the task with the latency
688 * @cpu: The cpu that initiated the trace.
689 *
690 * Flip the buffers between the @tr and the max_tr and record information
691 * about which task was the cause of this latency.
692 */
693void
694update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
695{
696 struct ring_buffer *buf = tr->buffer;
697
698 if (trace_stop_count)
699 return;
700
701 WARN_ON_ONCE(!irqs_disabled());
702 if (!current_trace->use_max_tr) {
703 WARN_ON_ONCE(1);
704 return;
705 }
706 arch_spin_lock(&ftrace_max_lock);
707
708 tr->buffer = max_tr.buffer;
709 max_tr.buffer = buf;
710
711 __update_max_tr(tr, tsk, cpu);
712 arch_spin_unlock(&ftrace_max_lock);
713}
714
715/**
716 * update_max_tr_single - only copy one trace over, and reset the rest
717 * @tr - tracer
718 * @tsk - task with the latency
719 * @cpu - the cpu of the buffer to copy.
720 *
721 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
722 */
723void
724update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
725{
726 int ret;
727
728 if (trace_stop_count)
729 return;
730
731 WARN_ON_ONCE(!irqs_disabled());
732 if (!current_trace->use_max_tr) {
733 WARN_ON_ONCE(1);
734 return;
735 }
736
737 arch_spin_lock(&ftrace_max_lock);
738
739 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
740
741 if (ret == -EBUSY) {
742 /*
743 * We failed to swap the buffer due to a commit taking
744 * place on this CPU. We fail to record, but we reset
745 * the max trace buffer (no one writes directly to it)
746 * and flag that it failed.
747 */
748 trace_array_printk(&max_tr, _THIS_IP_,
749 "Failed to swap buffers due to commit in progress\n");
750 }
751
752 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
753
754 __update_max_tr(tr, tsk, cpu);
755 arch_spin_unlock(&ftrace_max_lock);
756}
757#endif /* CONFIG_TRACER_MAX_TRACE */
758
759/**
760 * register_tracer - register a tracer with the ftrace system.
761 * @type - the plugin for the tracer
762 *
763 * Register a new plugin tracer.
764 */
765int register_tracer(struct tracer *type)
766{
767 struct tracer *t;
768 int ret = 0;
769
770 if (!type->name) {
771 pr_info("Tracer must have a name\n");
772 return -1;
773 }
774
775 if (strlen(type->name) >= MAX_TRACER_SIZE) {
776 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
777 return -1;
778 }
779
780 mutex_lock(&trace_types_lock);
781
782 tracing_selftest_running = true;
783
784 for (t = trace_types; t; t = t->next) {
785 if (strcmp(type->name, t->name) == 0) {
786 /* already found */
787 pr_info("Tracer %s already registered\n",
788 type->name);
789 ret = -1;
790 goto out;
791 }
792 }
793
794 if (!type->set_flag)
795 type->set_flag = &dummy_set_flag;
796 if (!type->flags)
797 type->flags = &dummy_tracer_flags;
798 else
799 if (!type->flags->opts)
800 type->flags->opts = dummy_tracer_opt;
801 if (!type->wait_pipe)
802 type->wait_pipe = default_wait_pipe;
803
804
805#ifdef CONFIG_FTRACE_STARTUP_TEST
806 if (type->selftest && !tracing_selftest_disabled) {
807 struct tracer *saved_tracer = current_trace;
808 struct trace_array *tr = &global_trace;
809
810 /*
811 * Run a selftest on this tracer.
812 * Here we reset the trace buffer, and set the current
813 * tracer to be this tracer. The tracer can then run some
814 * internal tracing to verify that everything is in order.
815 * If we fail, we do not register this tracer.
816 */
817 tracing_reset_online_cpus(tr);
818
819 current_trace = type;
820
821 /* If we expanded the buffers, make sure the max is expanded too */
822 if (ring_buffer_expanded && type->use_max_tr)
823 ring_buffer_resize(max_tr.buffer, trace_buf_size,
824 RING_BUFFER_ALL_CPUS);
825
826 /* the test is responsible for initializing and enabling */
827 pr_info("Testing tracer %s: ", type->name);
828 ret = type->selftest(type, tr);
829 /* the test is responsible for resetting too */
830 current_trace = saved_tracer;
831 if (ret) {
832 printk(KERN_CONT "FAILED!\n");
833 goto out;
834 }
835 /* Only reset on passing, to avoid touching corrupted buffers */
836 tracing_reset_online_cpus(tr);
837
838 /* Shrink the max buffer again */
839 if (ring_buffer_expanded && type->use_max_tr)
840 ring_buffer_resize(max_tr.buffer, 1,
841 RING_BUFFER_ALL_CPUS);
842
843 printk(KERN_CONT "PASSED\n");
844 }
845#endif
846
847 type->next = trace_types;
848 trace_types = type;
849
850 out:
851 tracing_selftest_running = false;
852 mutex_unlock(&trace_types_lock);
853
854 if (ret || !default_bootup_tracer)
855 goto out_unlock;
856
857 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
858 goto out_unlock;
859
860 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
861 /* Do we want this tracer to start on bootup? */
862 tracing_set_tracer(type->name);
863 default_bootup_tracer = NULL;
864 /* disable other selftests, since this will break it. */
865 tracing_selftest_disabled = 1;
866#ifdef CONFIG_FTRACE_STARTUP_TEST
867 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
868 type->name);
869#endif
870
871 out_unlock:
872 return ret;
873}
874
875void unregister_tracer(struct tracer *type)
876{
877 struct tracer **t;
878
879 mutex_lock(&trace_types_lock);
880 for (t = &trace_types; *t; t = &(*t)->next) {
881 if (*t == type)
882 goto found;
883 }
884 pr_info("Tracer %s not registered\n", type->name);
885 goto out;
886
887 found:
888 *t = (*t)->next;
889
890 if (type == current_trace && tracer_enabled) {
891 tracer_enabled = 0;
892 tracing_stop();
893 if (current_trace->stop)
894 current_trace->stop(&global_trace);
895 current_trace = &nop_trace;
896 }
897out:
898 mutex_unlock(&trace_types_lock);
899}
900
901void tracing_reset(struct trace_array *tr, int cpu)
902{
903 struct ring_buffer *buffer = tr->buffer;
904
905 ring_buffer_record_disable(buffer);
906
907 /* Make sure all commits have finished */
908 synchronize_sched();
909 ring_buffer_reset_cpu(buffer, cpu);
910
911 ring_buffer_record_enable(buffer);
912}
913
914void tracing_reset_online_cpus(struct trace_array *tr)
915{
916 struct ring_buffer *buffer = tr->buffer;
917 int cpu;
918
919 ring_buffer_record_disable(buffer);
920
921 /* Make sure all commits have finished */
922 synchronize_sched();
923
924 tr->time_start = ftrace_now(tr->cpu);
925
926 for_each_online_cpu(cpu)
927 ring_buffer_reset_cpu(buffer, cpu);
928
929 ring_buffer_record_enable(buffer);
930}
931
932void tracing_reset_current(int cpu)
933{
934 tracing_reset(&global_trace, cpu);
935}
936
937void tracing_reset_current_online_cpus(void)
938{
939 tracing_reset_online_cpus(&global_trace);
940}
941
942#define SAVED_CMDLINES 128
943#define NO_CMDLINE_MAP UINT_MAX
944static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
945static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
946static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
947static int cmdline_idx;
948static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
949
950/* temporary disable recording */
951static atomic_t trace_record_cmdline_disabled __read_mostly;
952
953static void trace_init_cmdlines(void)
954{
955 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
956 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
957 cmdline_idx = 0;
958}
959
960int is_tracing_stopped(void)
961{
962 return trace_stop_count;
963}
964
965/**
966 * ftrace_off_permanent - disable all ftrace code permanently
967 *
968 * This should only be called when a serious anomally has
969 * been detected. This will turn off the function tracing,
970 * ring buffers, and other tracing utilites. It takes no
971 * locks and can be called from any context.
972 */
973void ftrace_off_permanent(void)
974{
975 tracing_disabled = 1;
976 ftrace_stop();
977 tracing_off_permanent();
978}
979
980/**
981 * tracing_start - quick start of the tracer
982 *
983 * If tracing is enabled but was stopped by tracing_stop,
984 * this will start the tracer back up.
985 */
986void tracing_start(void)
987{
988 struct ring_buffer *buffer;
989 unsigned long flags;
990
991 if (tracing_disabled)
992 return;
993
994 raw_spin_lock_irqsave(&tracing_start_lock, flags);
995 if (--trace_stop_count) {
996 if (trace_stop_count < 0) {
997 /* Someone screwed up their debugging */
998 WARN_ON_ONCE(1);
999 trace_stop_count = 0;
1000 }
1001 goto out;
1002 }
1003
1004 /* Prevent the buffers from switching */
1005 arch_spin_lock(&ftrace_max_lock);
1006
1007 buffer = global_trace.buffer;
1008 if (buffer)
1009 ring_buffer_record_enable(buffer);
1010
1011 buffer = max_tr.buffer;
1012 if (buffer)
1013 ring_buffer_record_enable(buffer);
1014
1015 arch_spin_unlock(&ftrace_max_lock);
1016
1017 ftrace_start();
1018 out:
1019 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1020}
1021
1022/**
1023 * tracing_stop - quick stop of the tracer
1024 *
1025 * Light weight way to stop tracing. Use in conjunction with
1026 * tracing_start.
1027 */
1028void tracing_stop(void)
1029{
1030 struct ring_buffer *buffer;
1031 unsigned long flags;
1032
1033 ftrace_stop();
1034 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1035 if (trace_stop_count++)
1036 goto out;
1037
1038 /* Prevent the buffers from switching */
1039 arch_spin_lock(&ftrace_max_lock);
1040
1041 buffer = global_trace.buffer;
1042 if (buffer)
1043 ring_buffer_record_disable(buffer);
1044
1045 buffer = max_tr.buffer;
1046 if (buffer)
1047 ring_buffer_record_disable(buffer);
1048
1049 arch_spin_unlock(&ftrace_max_lock);
1050
1051 out:
1052 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1053}
1054
1055void trace_stop_cmdline_recording(void);
1056
1057static void trace_save_cmdline(struct task_struct *tsk)
1058{
1059 unsigned pid, idx;
1060
1061 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1062 return;
1063
1064 /*
1065 * It's not the end of the world if we don't get
1066 * the lock, but we also don't want to spin
1067 * nor do we want to disable interrupts,
1068 * so if we miss here, then better luck next time.
1069 */
1070 if (!arch_spin_trylock(&trace_cmdline_lock))
1071 return;
1072
1073 idx = map_pid_to_cmdline[tsk->pid];
1074 if (idx == NO_CMDLINE_MAP) {
1075 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1076
1077 /*
1078 * Check whether the cmdline buffer at idx has a pid
1079 * mapped. We are going to overwrite that entry so we
1080 * need to clear the map_pid_to_cmdline. Otherwise we
1081 * would read the new comm for the old pid.
1082 */
1083 pid = map_cmdline_to_pid[idx];
1084 if (pid != NO_CMDLINE_MAP)
1085 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1086
1087 map_cmdline_to_pid[idx] = tsk->pid;
1088 map_pid_to_cmdline[tsk->pid] = idx;
1089
1090 cmdline_idx = idx;
1091 }
1092
1093 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1094
1095 arch_spin_unlock(&trace_cmdline_lock);
1096}
1097
1098void trace_find_cmdline(int pid, char comm[])
1099{
1100 unsigned map;
1101
1102 if (!pid) {
1103 strcpy(comm, "<idle>");
1104 return;
1105 }
1106
1107 if (WARN_ON_ONCE(pid < 0)) {
1108 strcpy(comm, "<XXX>");
1109 return;
1110 }
1111
1112 if (pid > PID_MAX_DEFAULT) {
1113 strcpy(comm, "<...>");
1114 return;
1115 }
1116
1117 preempt_disable();
1118 arch_spin_lock(&trace_cmdline_lock);
1119 map = map_pid_to_cmdline[pid];
1120 if (map != NO_CMDLINE_MAP)
1121 strcpy(comm, saved_cmdlines[map]);
1122 else
1123 strcpy(comm, "<...>");
1124
1125 arch_spin_unlock(&trace_cmdline_lock);
1126 preempt_enable();
1127}
1128
1129void tracing_record_cmdline(struct task_struct *tsk)
1130{
1131 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1132 !tracing_is_on())
1133 return;
1134
1135 trace_save_cmdline(tsk);
1136}
1137
1138void
1139tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1140 int pc)
1141{
1142 struct task_struct *tsk = current;
1143
1144 entry->preempt_count = pc & 0xff;
1145 entry->pid = (tsk) ? tsk->pid : 0;
1146 entry->padding = 0;
1147 entry->flags =
1148#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1149 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1150#else
1151 TRACE_FLAG_IRQS_NOSUPPORT |
1152#endif
1153 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1154 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1155 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1156}
1157EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1158
1159struct ring_buffer_event *
1160trace_buffer_lock_reserve(struct ring_buffer *buffer,
1161 int type,
1162 unsigned long len,
1163 unsigned long flags, int pc)
1164{
1165 struct ring_buffer_event *event;
1166
1167 event = ring_buffer_lock_reserve(buffer, len);
1168 if (event != NULL) {
1169 struct trace_entry *ent = ring_buffer_event_data(event);
1170
1171 tracing_generic_entry_update(ent, flags, pc);
1172 ent->type = type;
1173 }
1174
1175 return event;
1176}
1177
1178static inline void
1179__trace_buffer_unlock_commit(struct ring_buffer *buffer,
1180 struct ring_buffer_event *event,
1181 unsigned long flags, int pc,
1182 int wake)
1183{
1184 ring_buffer_unlock_commit(buffer, event);
1185
1186 ftrace_trace_stack(buffer, flags, 6, pc);
1187 ftrace_trace_userstack(buffer, flags, pc);
1188
1189 if (wake)
1190 trace_wake_up();
1191}
1192
1193void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1194 struct ring_buffer_event *event,
1195 unsigned long flags, int pc)
1196{
1197 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1198}
1199
1200struct ring_buffer_event *
1201trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1202 int type, unsigned long len,
1203 unsigned long flags, int pc)
1204{
1205 *current_rb = global_trace.buffer;
1206 return trace_buffer_lock_reserve(*current_rb,
1207 type, len, flags, pc);
1208}
1209EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1210
1211void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1212 struct ring_buffer_event *event,
1213 unsigned long flags, int pc)
1214{
1215 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1216}
1217EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1218
1219void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1220 struct ring_buffer_event *event,
1221 unsigned long flags, int pc)
1222{
1223 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1224}
1225EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1226
1227void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1228 struct ring_buffer_event *event,
1229 unsigned long flags, int pc,
1230 struct pt_regs *regs)
1231{
1232 ring_buffer_unlock_commit(buffer, event);
1233
1234 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1235 ftrace_trace_userstack(buffer, flags, pc);
1236}
1237EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1238
1239void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1240 struct ring_buffer_event *event)
1241{
1242 ring_buffer_discard_commit(buffer, event);
1243}
1244EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1245
1246void
1247trace_function(struct trace_array *tr,
1248 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1249 int pc)
1250{
1251 struct ftrace_event_call *call = &event_function;
1252 struct ring_buffer *buffer = tr->buffer;
1253 struct ring_buffer_event *event;
1254 struct ftrace_entry *entry;
1255
1256 /* If we are reading the ring buffer, don't trace */
1257 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1258 return;
1259
1260 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1261 flags, pc);
1262 if (!event)
1263 return;
1264 entry = ring_buffer_event_data(event);
1265 entry->ip = ip;
1266 entry->parent_ip = parent_ip;
1267
1268 if (!filter_check_discard(call, entry, buffer, event))
1269 ring_buffer_unlock_commit(buffer, event);
1270}
1271
1272void
1273ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1274 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1275 int pc)
1276{
1277 if (likely(!atomic_read(&data->disabled)))
1278 trace_function(tr, ip, parent_ip, flags, pc);
1279}
1280
1281#ifdef CONFIG_STACKTRACE
1282
1283#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1284struct ftrace_stack {
1285 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1286};
1287
1288static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1289static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1290
1291static void __ftrace_trace_stack(struct ring_buffer *buffer,
1292 unsigned long flags,
1293 int skip, int pc, struct pt_regs *regs)
1294{
1295 struct ftrace_event_call *call = &event_kernel_stack;
1296 struct ring_buffer_event *event;
1297 struct stack_entry *entry;
1298 struct stack_trace trace;
1299 int use_stack;
1300 int size = FTRACE_STACK_ENTRIES;
1301
1302 trace.nr_entries = 0;
1303 trace.skip = skip;
1304
1305 /*
1306 * Since events can happen in NMIs there's no safe way to
1307 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1308 * or NMI comes in, it will just have to use the default
1309 * FTRACE_STACK_SIZE.
1310 */
1311 preempt_disable_notrace();
1312
1313 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1314 /*
1315 * We don't need any atomic variables, just a barrier.
1316 * If an interrupt comes in, we don't care, because it would
1317 * have exited and put the counter back to what we want.
1318 * We just need a barrier to keep gcc from moving things
1319 * around.
1320 */
1321 barrier();
1322 if (use_stack == 1) {
1323 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1324 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1325
1326 if (regs)
1327 save_stack_trace_regs(regs, &trace);
1328 else
1329 save_stack_trace(&trace);
1330
1331 if (trace.nr_entries > size)
1332 size = trace.nr_entries;
1333 } else
1334 /* From now on, use_stack is a boolean */
1335 use_stack = 0;
1336
1337 size *= sizeof(unsigned long);
1338
1339 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1340 sizeof(*entry) + size, flags, pc);
1341 if (!event)
1342 goto out;
1343 entry = ring_buffer_event_data(event);
1344
1345 memset(&entry->caller, 0, size);
1346
1347 if (use_stack)
1348 memcpy(&entry->caller, trace.entries,
1349 trace.nr_entries * sizeof(unsigned long));
1350 else {
1351 trace.max_entries = FTRACE_STACK_ENTRIES;
1352 trace.entries = entry->caller;
1353 if (regs)
1354 save_stack_trace_regs(regs, &trace);
1355 else
1356 save_stack_trace(&trace);
1357 }
1358
1359 entry->size = trace.nr_entries;
1360
1361 if (!filter_check_discard(call, entry, buffer, event))
1362 ring_buffer_unlock_commit(buffer, event);
1363
1364 out:
1365 /* Again, don't let gcc optimize things here */
1366 barrier();
1367 __get_cpu_var(ftrace_stack_reserve)--;
1368 preempt_enable_notrace();
1369
1370}
1371
1372void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1373 int skip, int pc, struct pt_regs *regs)
1374{
1375 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1376 return;
1377
1378 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1379}
1380
1381void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1382 int skip, int pc)
1383{
1384 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1385 return;
1386
1387 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1388}
1389
1390void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1391 int pc)
1392{
1393 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1394}
1395
1396/**
1397 * trace_dump_stack - record a stack back trace in the trace buffer
1398 */
1399void trace_dump_stack(void)
1400{
1401 unsigned long flags;
1402
1403 if (tracing_disabled || tracing_selftest_running)
1404 return;
1405
1406 local_save_flags(flags);
1407
1408 /* skipping 3 traces, seems to get us at the caller of this function */
1409 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1410}
1411
1412static DEFINE_PER_CPU(int, user_stack_count);
1413
1414void
1415ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1416{
1417 struct ftrace_event_call *call = &event_user_stack;
1418 struct ring_buffer_event *event;
1419 struct userstack_entry *entry;
1420 struct stack_trace trace;
1421
1422 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1423 return;
1424
1425 /*
1426 * NMIs can not handle page faults, even with fix ups.
1427 * The save user stack can (and often does) fault.
1428 */
1429 if (unlikely(in_nmi()))
1430 return;
1431
1432 /*
1433 * prevent recursion, since the user stack tracing may
1434 * trigger other kernel events.
1435 */
1436 preempt_disable();
1437 if (__this_cpu_read(user_stack_count))
1438 goto out;
1439
1440 __this_cpu_inc(user_stack_count);
1441
1442 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1443 sizeof(*entry), flags, pc);
1444 if (!event)
1445 goto out_drop_count;
1446 entry = ring_buffer_event_data(event);
1447
1448 entry->tgid = current->tgid;
1449 memset(&entry->caller, 0, sizeof(entry->caller));
1450
1451 trace.nr_entries = 0;
1452 trace.max_entries = FTRACE_STACK_ENTRIES;
1453 trace.skip = 0;
1454 trace.entries = entry->caller;
1455
1456 save_stack_trace_user(&trace);
1457 if (!filter_check_discard(call, entry, buffer, event))
1458 ring_buffer_unlock_commit(buffer, event);
1459
1460 out_drop_count:
1461 __this_cpu_dec(user_stack_count);
1462 out:
1463 preempt_enable();
1464}
1465
1466#ifdef UNUSED
1467static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1468{
1469 ftrace_trace_userstack(tr, flags, preempt_count());
1470}
1471#endif /* UNUSED */
1472
1473#endif /* CONFIG_STACKTRACE */
1474
1475/* created for use with alloc_percpu */
1476struct trace_buffer_struct {
1477 char buffer[TRACE_BUF_SIZE];
1478};
1479
1480static struct trace_buffer_struct *trace_percpu_buffer;
1481static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1482static struct trace_buffer_struct *trace_percpu_irq_buffer;
1483static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1484
1485/*
1486 * The buffer used is dependent on the context. There is a per cpu
1487 * buffer for normal context, softirq contex, hard irq context and
1488 * for NMI context. Thise allows for lockless recording.
1489 *
1490 * Note, if the buffers failed to be allocated, then this returns NULL
1491 */
1492static char *get_trace_buf(void)
1493{
1494 struct trace_buffer_struct *percpu_buffer;
1495 struct trace_buffer_struct *buffer;
1496
1497 /*
1498 * If we have allocated per cpu buffers, then we do not
1499 * need to do any locking.
1500 */
1501 if (in_nmi())
1502 percpu_buffer = trace_percpu_nmi_buffer;
1503 else if (in_irq())
1504 percpu_buffer = trace_percpu_irq_buffer;
1505 else if (in_softirq())
1506 percpu_buffer = trace_percpu_sirq_buffer;
1507 else
1508 percpu_buffer = trace_percpu_buffer;
1509
1510 if (!percpu_buffer)
1511 return NULL;
1512
1513 buffer = per_cpu_ptr(percpu_buffer, smp_processor_id());
1514
1515 return buffer->buffer;
1516}
1517
1518static int alloc_percpu_trace_buffer(void)
1519{
1520 struct trace_buffer_struct *buffers;
1521 struct trace_buffer_struct *sirq_buffers;
1522 struct trace_buffer_struct *irq_buffers;
1523 struct trace_buffer_struct *nmi_buffers;
1524
1525 buffers = alloc_percpu(struct trace_buffer_struct);
1526 if (!buffers)
1527 goto err_warn;
1528
1529 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1530 if (!sirq_buffers)
1531 goto err_sirq;
1532
1533 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1534 if (!irq_buffers)
1535 goto err_irq;
1536
1537 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1538 if (!nmi_buffers)
1539 goto err_nmi;
1540
1541 trace_percpu_buffer = buffers;
1542 trace_percpu_sirq_buffer = sirq_buffers;
1543 trace_percpu_irq_buffer = irq_buffers;
1544 trace_percpu_nmi_buffer = nmi_buffers;
1545
1546 return 0;
1547
1548 err_nmi:
1549 free_percpu(irq_buffers);
1550 err_irq:
1551 free_percpu(sirq_buffers);
1552 err_sirq:
1553 free_percpu(buffers);
1554 err_warn:
1555 WARN(1, "Could not allocate percpu trace_printk buffer");
1556 return -ENOMEM;
1557}
1558
1559void trace_printk_init_buffers(void)
1560{
1561 static int buffers_allocated;
1562
1563 if (buffers_allocated)
1564 return;
1565
1566 if (alloc_percpu_trace_buffer())
1567 return;
1568
1569 pr_info("ftrace: Allocated trace_printk buffers\n");
1570
1571 buffers_allocated = 1;
1572}
1573
1574/**
1575 * trace_vbprintk - write binary msg to tracing buffer
1576 *
1577 */
1578int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1579{
1580 struct ftrace_event_call *call = &event_bprint;
1581 struct ring_buffer_event *event;
1582 struct ring_buffer *buffer;
1583 struct trace_array *tr = &global_trace;
1584 struct bprint_entry *entry;
1585 unsigned long flags;
1586 char *tbuffer;
1587 int len = 0, size, pc;
1588
1589 if (unlikely(tracing_selftest_running || tracing_disabled))
1590 return 0;
1591
1592 /* Don't pollute graph traces with trace_vprintk internals */
1593 pause_graph_tracing();
1594
1595 pc = preempt_count();
1596 preempt_disable_notrace();
1597
1598 tbuffer = get_trace_buf();
1599 if (!tbuffer) {
1600 len = 0;
1601 goto out;
1602 }
1603
1604 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1605
1606 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1607 goto out;
1608
1609 local_save_flags(flags);
1610 size = sizeof(*entry) + sizeof(u32) * len;
1611 buffer = tr->buffer;
1612 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1613 flags, pc);
1614 if (!event)
1615 goto out;
1616 entry = ring_buffer_event_data(event);
1617 entry->ip = ip;
1618 entry->fmt = fmt;
1619
1620 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1621 if (!filter_check_discard(call, entry, buffer, event)) {
1622 ring_buffer_unlock_commit(buffer, event);
1623 ftrace_trace_stack(buffer, flags, 6, pc);
1624 }
1625
1626out:
1627 preempt_enable_notrace();
1628 unpause_graph_tracing();
1629
1630 return len;
1631}
1632EXPORT_SYMBOL_GPL(trace_vbprintk);
1633
1634int trace_array_printk(struct trace_array *tr,
1635 unsigned long ip, const char *fmt, ...)
1636{
1637 int ret;
1638 va_list ap;
1639
1640 if (!(trace_flags & TRACE_ITER_PRINTK))
1641 return 0;
1642
1643 va_start(ap, fmt);
1644 ret = trace_array_vprintk(tr, ip, fmt, ap);
1645 va_end(ap);
1646 return ret;
1647}
1648
1649int trace_array_vprintk(struct trace_array *tr,
1650 unsigned long ip, const char *fmt, va_list args)
1651{
1652 struct ftrace_event_call *call = &event_print;
1653 struct ring_buffer_event *event;
1654 struct ring_buffer *buffer;
1655 int len = 0, size, pc;
1656 struct print_entry *entry;
1657 unsigned long flags;
1658 char *tbuffer;
1659
1660 if (tracing_disabled || tracing_selftest_running)
1661 return 0;
1662
1663 /* Don't pollute graph traces with trace_vprintk internals */
1664 pause_graph_tracing();
1665
1666 pc = preempt_count();
1667 preempt_disable_notrace();
1668
1669
1670 tbuffer = get_trace_buf();
1671 if (!tbuffer) {
1672 len = 0;
1673 goto out;
1674 }
1675
1676 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1677 if (len > TRACE_BUF_SIZE)
1678 goto out;
1679
1680 local_save_flags(flags);
1681 size = sizeof(*entry) + len + 1;
1682 buffer = tr->buffer;
1683 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1684 flags, pc);
1685 if (!event)
1686 goto out;
1687 entry = ring_buffer_event_data(event);
1688 entry->ip = ip;
1689
1690 memcpy(&entry->buf, tbuffer, len);
1691 entry->buf[len] = '\0';
1692 if (!filter_check_discard(call, entry, buffer, event)) {
1693 ring_buffer_unlock_commit(buffer, event);
1694 ftrace_trace_stack(buffer, flags, 6, pc);
1695 }
1696 out:
1697 preempt_enable_notrace();
1698 unpause_graph_tracing();
1699
1700 return len;
1701}
1702
1703int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1704{
1705 return trace_array_vprintk(&global_trace, ip, fmt, args);
1706}
1707EXPORT_SYMBOL_GPL(trace_vprintk);
1708
1709static void trace_iterator_increment(struct trace_iterator *iter)
1710{
1711 iter->idx++;
1712 if (iter->buffer_iter[iter->cpu])
1713 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1714}
1715
1716static struct trace_entry *
1717peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1718 unsigned long *lost_events)
1719{
1720 struct ring_buffer_event *event;
1721 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1722
1723 if (buf_iter)
1724 event = ring_buffer_iter_peek(buf_iter, ts);
1725 else
1726 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1727 lost_events);
1728
1729 if (event) {
1730 iter->ent_size = ring_buffer_event_length(event);
1731 return ring_buffer_event_data(event);
1732 }
1733 iter->ent_size = 0;
1734 return NULL;
1735}
1736
1737static struct trace_entry *
1738__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1739 unsigned long *missing_events, u64 *ent_ts)
1740{
1741 struct ring_buffer *buffer = iter->tr->buffer;
1742 struct trace_entry *ent, *next = NULL;
1743 unsigned long lost_events = 0, next_lost = 0;
1744 int cpu_file = iter->cpu_file;
1745 u64 next_ts = 0, ts;
1746 int next_cpu = -1;
1747 int next_size = 0;
1748 int cpu;
1749
1750 /*
1751 * If we are in a per_cpu trace file, don't bother by iterating over
1752 * all cpu and peek directly.
1753 */
1754 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1755 if (ring_buffer_empty_cpu(buffer, cpu_file))
1756 return NULL;
1757 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1758 if (ent_cpu)
1759 *ent_cpu = cpu_file;
1760
1761 return ent;
1762 }
1763
1764 for_each_tracing_cpu(cpu) {
1765
1766 if (ring_buffer_empty_cpu(buffer, cpu))
1767 continue;
1768
1769 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1770
1771 /*
1772 * Pick the entry with the smallest timestamp:
1773 */
1774 if (ent && (!next || ts < next_ts)) {
1775 next = ent;
1776 next_cpu = cpu;
1777 next_ts = ts;
1778 next_lost = lost_events;
1779 next_size = iter->ent_size;
1780 }
1781 }
1782
1783 iter->ent_size = next_size;
1784
1785 if (ent_cpu)
1786 *ent_cpu = next_cpu;
1787
1788 if (ent_ts)
1789 *ent_ts = next_ts;
1790
1791 if (missing_events)
1792 *missing_events = next_lost;
1793
1794 return next;
1795}
1796
1797/* Find the next real entry, without updating the iterator itself */
1798struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1799 int *ent_cpu, u64 *ent_ts)
1800{
1801 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1802}
1803
1804/* Find the next real entry, and increment the iterator to the next entry */
1805void *trace_find_next_entry_inc(struct trace_iterator *iter)
1806{
1807 iter->ent = __find_next_entry(iter, &iter->cpu,
1808 &iter->lost_events, &iter->ts);
1809
1810 if (iter->ent)
1811 trace_iterator_increment(iter);
1812
1813 return iter->ent ? iter : NULL;
1814}
1815
1816static void trace_consume(struct trace_iterator *iter)
1817{
1818 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1819 &iter->lost_events);
1820}
1821
1822static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1823{
1824 struct trace_iterator *iter = m->private;
1825 int i = (int)*pos;
1826 void *ent;
1827
1828 WARN_ON_ONCE(iter->leftover);
1829
1830 (*pos)++;
1831
1832 /* can't go backwards */
1833 if (iter->idx > i)
1834 return NULL;
1835
1836 if (iter->idx < 0)
1837 ent = trace_find_next_entry_inc(iter);
1838 else
1839 ent = iter;
1840
1841 while (ent && iter->idx < i)
1842 ent = trace_find_next_entry_inc(iter);
1843
1844 iter->pos = *pos;
1845
1846 return ent;
1847}
1848
1849void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1850{
1851 struct trace_array *tr = iter->tr;
1852 struct ring_buffer_event *event;
1853 struct ring_buffer_iter *buf_iter;
1854 unsigned long entries = 0;
1855 u64 ts;
1856
1857 tr->data[cpu]->skipped_entries = 0;
1858
1859 if (!iter->buffer_iter[cpu])
1860 return;
1861
1862 buf_iter = iter->buffer_iter[cpu];
1863 ring_buffer_iter_reset(buf_iter);
1864
1865 /*
1866 * We could have the case with the max latency tracers
1867 * that a reset never took place on a cpu. This is evident
1868 * by the timestamp being before the start of the buffer.
1869 */
1870 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1871 if (ts >= iter->tr->time_start)
1872 break;
1873 entries++;
1874 ring_buffer_read(buf_iter, NULL);
1875 }
1876
1877 tr->data[cpu]->skipped_entries = entries;
1878}
1879
1880/*
1881 * The current tracer is copied to avoid a global locking
1882 * all around.
1883 */
1884static void *s_start(struct seq_file *m, loff_t *pos)
1885{
1886 struct trace_iterator *iter = m->private;
1887 static struct tracer *old_tracer;
1888 int cpu_file = iter->cpu_file;
1889 void *p = NULL;
1890 loff_t l = 0;
1891 int cpu;
1892
1893 /* copy the tracer to avoid using a global lock all around */
1894 mutex_lock(&trace_types_lock);
1895 if (unlikely(old_tracer != current_trace && current_trace)) {
1896 old_tracer = current_trace;
1897 *iter->trace = *current_trace;
1898 }
1899 mutex_unlock(&trace_types_lock);
1900
1901 atomic_inc(&trace_record_cmdline_disabled);
1902
1903 if (*pos != iter->pos) {
1904 iter->ent = NULL;
1905 iter->cpu = 0;
1906 iter->idx = -1;
1907
1908 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1909 for_each_tracing_cpu(cpu)
1910 tracing_iter_reset(iter, cpu);
1911 } else
1912 tracing_iter_reset(iter, cpu_file);
1913
1914 iter->leftover = 0;
1915 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1916 ;
1917
1918 } else {
1919 /*
1920 * If we overflowed the seq_file before, then we want
1921 * to just reuse the trace_seq buffer again.
1922 */
1923 if (iter->leftover)
1924 p = iter;
1925 else {
1926 l = *pos - 1;
1927 p = s_next(m, p, &l);
1928 }
1929 }
1930
1931 trace_event_read_lock();
1932 trace_access_lock(cpu_file);
1933 return p;
1934}
1935
1936static void s_stop(struct seq_file *m, void *p)
1937{
1938 struct trace_iterator *iter = m->private;
1939
1940 atomic_dec(&trace_record_cmdline_disabled);
1941 trace_access_unlock(iter->cpu_file);
1942 trace_event_read_unlock();
1943}
1944
1945static void
1946get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1947{
1948 unsigned long count;
1949 int cpu;
1950
1951 *total = 0;
1952 *entries = 0;
1953
1954 for_each_tracing_cpu(cpu) {
1955 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1956 /*
1957 * If this buffer has skipped entries, then we hold all
1958 * entries for the trace and we need to ignore the
1959 * ones before the time stamp.
1960 */
1961 if (tr->data[cpu]->skipped_entries) {
1962 count -= tr->data[cpu]->skipped_entries;
1963 /* total is the same as the entries */
1964 *total += count;
1965 } else
1966 *total += count +
1967 ring_buffer_overrun_cpu(tr->buffer, cpu);
1968 *entries += count;
1969 }
1970}
1971
1972static void print_lat_help_header(struct seq_file *m)
1973{
1974 seq_puts(m, "# _------=> CPU# \n");
1975 seq_puts(m, "# / _-----=> irqs-off \n");
1976 seq_puts(m, "# | / _----=> need-resched \n");
1977 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1978 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1979 seq_puts(m, "# |||| / delay \n");
1980 seq_puts(m, "# cmd pid ||||| time | caller \n");
1981 seq_puts(m, "# \\ / ||||| \\ | / \n");
1982}
1983
1984static void print_event_info(struct trace_array *tr, struct seq_file *m)
1985{
1986 unsigned long total;
1987 unsigned long entries;
1988
1989 get_total_entries(tr, &total, &entries);
1990 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1991 entries, total, num_online_cpus());
1992 seq_puts(m, "#\n");
1993}
1994
1995static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
1996{
1997 print_event_info(tr, m);
1998 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1999 seq_puts(m, "# | | | | |\n");
2000}
2001
2002static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2003{
2004 print_event_info(tr, m);
2005 seq_puts(m, "# _-----=> irqs-off\n");
2006 seq_puts(m, "# / _----=> need-resched\n");
2007 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2008 seq_puts(m, "# || / _--=> preempt-depth\n");
2009 seq_puts(m, "# ||| / delay\n");
2010 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2011 seq_puts(m, "# | | | |||| | |\n");
2012}
2013
2014void
2015print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2016{
2017 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2018 struct trace_array *tr = iter->tr;
2019 struct trace_array_cpu *data = tr->data[tr->cpu];
2020 struct tracer *type = current_trace;
2021 unsigned long entries;
2022 unsigned long total;
2023 const char *name = "preemption";
2024
2025 if (type)
2026 name = type->name;
2027
2028 get_total_entries(tr, &total, &entries);
2029
2030 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2031 name, UTS_RELEASE);
2032 seq_puts(m, "# -----------------------------------"
2033 "---------------------------------\n");
2034 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2035 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2036 nsecs_to_usecs(data->saved_latency),
2037 entries,
2038 total,
2039 tr->cpu,
2040#if defined(CONFIG_PREEMPT_NONE)
2041 "server",
2042#elif defined(CONFIG_PREEMPT_VOLUNTARY)
2043 "desktop",
2044#elif defined(CONFIG_PREEMPT)
2045 "preempt",
2046#else
2047 "unknown",
2048#endif
2049 /* These are reserved for later use */
2050 0, 0, 0, 0);
2051#ifdef CONFIG_SMP
2052 seq_printf(m, " #P:%d)\n", num_online_cpus());
2053#else
2054 seq_puts(m, ")\n");
2055#endif
2056 seq_puts(m, "# -----------------\n");
2057 seq_printf(m, "# | task: %.16s-%d "
2058 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2059 data->comm, data->pid, data->uid, data->nice,
2060 data->policy, data->rt_priority);
2061 seq_puts(m, "# -----------------\n");
2062
2063 if (data->critical_start) {
2064 seq_puts(m, "# => started at: ");
2065 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2066 trace_print_seq(m, &iter->seq);
2067 seq_puts(m, "\n# => ended at: ");
2068 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2069 trace_print_seq(m, &iter->seq);
2070 seq_puts(m, "\n#\n");
2071 }
2072
2073 seq_puts(m, "#\n");
2074}
2075
2076static void test_cpu_buff_start(struct trace_iterator *iter)
2077{
2078 struct trace_seq *s = &iter->seq;
2079
2080 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2081 return;
2082
2083 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2084 return;
2085
2086 if (cpumask_test_cpu(iter->cpu, iter->started))
2087 return;
2088
2089 if (iter->tr->data[iter->cpu]->skipped_entries)
2090 return;
2091
2092 cpumask_set_cpu(iter->cpu, iter->started);
2093
2094 /* Don't print started cpu buffer for the first entry of the trace */
2095 if (iter->idx > 1)
2096 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2097 iter->cpu);
2098}
2099
2100static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2101{
2102 struct trace_seq *s = &iter->seq;
2103 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2104 struct trace_entry *entry;
2105 struct trace_event *event;
2106
2107 entry = iter->ent;
2108
2109 test_cpu_buff_start(iter);
2110
2111 event = ftrace_find_event(entry->type);
2112
2113 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2114 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2115 if (!trace_print_lat_context(iter))
2116 goto partial;
2117 } else {
2118 if (!trace_print_context(iter))
2119 goto partial;
2120 }
2121 }
2122
2123 if (event)
2124 return event->funcs->trace(iter, sym_flags, event);
2125
2126 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2127 goto partial;
2128
2129 return TRACE_TYPE_HANDLED;
2130partial:
2131 return TRACE_TYPE_PARTIAL_LINE;
2132}
2133
2134static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2135{
2136 struct trace_seq *s = &iter->seq;
2137 struct trace_entry *entry;
2138 struct trace_event *event;
2139
2140 entry = iter->ent;
2141
2142 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2143 if (!trace_seq_printf(s, "%d %d %llu ",
2144 entry->pid, iter->cpu, iter->ts))
2145 goto partial;
2146 }
2147
2148 event = ftrace_find_event(entry->type);
2149 if (event)
2150 return event->funcs->raw(iter, 0, event);
2151
2152 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2153 goto partial;
2154
2155 return TRACE_TYPE_HANDLED;
2156partial:
2157 return TRACE_TYPE_PARTIAL_LINE;
2158}
2159
2160static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2161{
2162 struct trace_seq *s = &iter->seq;
2163 unsigned char newline = '\n';
2164 struct trace_entry *entry;
2165 struct trace_event *event;
2166
2167 entry = iter->ent;
2168
2169 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2170 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2171 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2172 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2173 }
2174
2175 event = ftrace_find_event(entry->type);
2176 if (event) {
2177 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2178 if (ret != TRACE_TYPE_HANDLED)
2179 return ret;
2180 }
2181
2182 SEQ_PUT_FIELD_RET(s, newline);
2183
2184 return TRACE_TYPE_HANDLED;
2185}
2186
2187static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2188{
2189 struct trace_seq *s = &iter->seq;
2190 struct trace_entry *entry;
2191 struct trace_event *event;
2192
2193 entry = iter->ent;
2194
2195 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2196 SEQ_PUT_FIELD_RET(s, entry->pid);
2197 SEQ_PUT_FIELD_RET(s, iter->cpu);
2198 SEQ_PUT_FIELD_RET(s, iter->ts);
2199 }
2200
2201 event = ftrace_find_event(entry->type);
2202 return event ? event->funcs->binary(iter, 0, event) :
2203 TRACE_TYPE_HANDLED;
2204}
2205
2206int trace_empty(struct trace_iterator *iter)
2207{
2208 int cpu;
2209
2210 /* If we are looking at one CPU buffer, only check that one */
2211 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2212 cpu = iter->cpu_file;
2213 if (iter->buffer_iter[cpu]) {
2214 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2215 return 0;
2216 } else {
2217 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2218 return 0;
2219 }
2220 return 1;
2221 }
2222
2223 for_each_tracing_cpu(cpu) {
2224 if (iter->buffer_iter[cpu]) {
2225 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2226 return 0;
2227 } else {
2228 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2229 return 0;
2230 }
2231 }
2232
2233 return 1;
2234}
2235
2236/* Called with trace_event_read_lock() held. */
2237enum print_line_t print_trace_line(struct trace_iterator *iter)
2238{
2239 enum print_line_t ret;
2240
2241 if (iter->lost_events &&
2242 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2243 iter->cpu, iter->lost_events))
2244 return TRACE_TYPE_PARTIAL_LINE;
2245
2246 if (iter->trace && iter->trace->print_line) {
2247 ret = iter->trace->print_line(iter);
2248 if (ret != TRACE_TYPE_UNHANDLED)
2249 return ret;
2250 }
2251
2252 if (iter->ent->type == TRACE_BPRINT &&
2253 trace_flags & TRACE_ITER_PRINTK &&
2254 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2255 return trace_print_bprintk_msg_only(iter);
2256
2257 if (iter->ent->type == TRACE_PRINT &&
2258 trace_flags & TRACE_ITER_PRINTK &&
2259 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2260 return trace_print_printk_msg_only(iter);
2261
2262 if (trace_flags & TRACE_ITER_BIN)
2263 return print_bin_fmt(iter);
2264
2265 if (trace_flags & TRACE_ITER_HEX)
2266 return print_hex_fmt(iter);
2267
2268 if (trace_flags & TRACE_ITER_RAW)
2269 return print_raw_fmt(iter);
2270
2271 return print_trace_fmt(iter);
2272}
2273
2274void trace_latency_header(struct seq_file *m)
2275{
2276 struct trace_iterator *iter = m->private;
2277
2278 /* print nothing if the buffers are empty */
2279 if (trace_empty(iter))
2280 return;
2281
2282 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2283 print_trace_header(m, iter);
2284
2285 if (!(trace_flags & TRACE_ITER_VERBOSE))
2286 print_lat_help_header(m);
2287}
2288
2289void trace_default_header(struct seq_file *m)
2290{
2291 struct trace_iterator *iter = m->private;
2292
2293 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2294 return;
2295
2296 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2297 /* print nothing if the buffers are empty */
2298 if (trace_empty(iter))
2299 return;
2300 print_trace_header(m, iter);
2301 if (!(trace_flags & TRACE_ITER_VERBOSE))
2302 print_lat_help_header(m);
2303 } else {
2304 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2305 if (trace_flags & TRACE_ITER_IRQ_INFO)
2306 print_func_help_header_irq(iter->tr, m);
2307 else
2308 print_func_help_header(iter->tr, m);
2309 }
2310 }
2311}
2312
2313static void test_ftrace_alive(struct seq_file *m)
2314{
2315 if (!ftrace_is_dead())
2316 return;
2317 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2318 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2319}
2320
2321static int s_show(struct seq_file *m, void *v)
2322{
2323 struct trace_iterator *iter = v;
2324 int ret;
2325
2326 if (iter->ent == NULL) {
2327 if (iter->tr) {
2328 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2329 seq_puts(m, "#\n");
2330 test_ftrace_alive(m);
2331 }
2332 if (iter->trace && iter->trace->print_header)
2333 iter->trace->print_header(m);
2334 else
2335 trace_default_header(m);
2336
2337 } else if (iter->leftover) {
2338 /*
2339 * If we filled the seq_file buffer earlier, we
2340 * want to just show it now.
2341 */
2342 ret = trace_print_seq(m, &iter->seq);
2343
2344 /* ret should this time be zero, but you never know */
2345 iter->leftover = ret;
2346
2347 } else {
2348 print_trace_line(iter);
2349 ret = trace_print_seq(m, &iter->seq);
2350 /*
2351 * If we overflow the seq_file buffer, then it will
2352 * ask us for this data again at start up.
2353 * Use that instead.
2354 * ret is 0 if seq_file write succeeded.
2355 * -1 otherwise.
2356 */
2357 iter->leftover = ret;
2358 }
2359
2360 return 0;
2361}
2362
2363static const struct seq_operations tracer_seq_ops = {
2364 .start = s_start,
2365 .next = s_next,
2366 .stop = s_stop,
2367 .show = s_show,
2368};
2369
2370static struct trace_iterator *
2371__tracing_open(struct inode *inode, struct file *file)
2372{
2373 long cpu_file = (long) inode->i_private;
2374 struct trace_iterator *iter;
2375 int cpu;
2376
2377 if (tracing_disabled)
2378 return ERR_PTR(-ENODEV);
2379
2380 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2381 if (!iter)
2382 return ERR_PTR(-ENOMEM);
2383
2384 /*
2385 * We make a copy of the current tracer to avoid concurrent
2386 * changes on it while we are reading.
2387 */
2388 mutex_lock(&trace_types_lock);
2389 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2390 if (!iter->trace)
2391 goto fail;
2392
2393 if (current_trace)
2394 *iter->trace = *current_trace;
2395
2396 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2397 goto fail;
2398
2399 if (current_trace && current_trace->print_max)
2400 iter->tr = &max_tr;
2401 else
2402 iter->tr = &global_trace;
2403 iter->pos = -1;
2404 mutex_init(&iter->mutex);
2405 iter->cpu_file = cpu_file;
2406
2407 /* Notify the tracer early; before we stop tracing. */
2408 if (iter->trace && iter->trace->open)
2409 iter->trace->open(iter);
2410
2411 /* Annotate start of buffers if we had overruns */
2412 if (ring_buffer_overruns(iter->tr->buffer))
2413 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2414
2415 /* stop the trace while dumping */
2416 tracing_stop();
2417
2418 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2419 for_each_tracing_cpu(cpu) {
2420 iter->buffer_iter[cpu] =
2421 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2422 }
2423 ring_buffer_read_prepare_sync();
2424 for_each_tracing_cpu(cpu) {
2425 ring_buffer_read_start(iter->buffer_iter[cpu]);
2426 tracing_iter_reset(iter, cpu);
2427 }
2428 } else {
2429 cpu = iter->cpu_file;
2430 iter->buffer_iter[cpu] =
2431 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2432 ring_buffer_read_prepare_sync();
2433 ring_buffer_read_start(iter->buffer_iter[cpu]);
2434 tracing_iter_reset(iter, cpu);
2435 }
2436
2437 mutex_unlock(&trace_types_lock);
2438
2439 return iter;
2440
2441 fail:
2442 mutex_unlock(&trace_types_lock);
2443 kfree(iter->trace);
2444 seq_release_private(inode, file);
2445 return ERR_PTR(-ENOMEM);
2446}
2447
2448int tracing_open_generic(struct inode *inode, struct file *filp)
2449{
2450 if (tracing_disabled)
2451 return -ENODEV;
2452
2453 filp->private_data = inode->i_private;
2454 return 0;
2455}
2456
2457static int tracing_release(struct inode *inode, struct file *file)
2458{
2459 struct seq_file *m = file->private_data;
2460 struct trace_iterator *iter;
2461 int cpu;
2462
2463 if (!(file->f_mode & FMODE_READ))
2464 return 0;
2465
2466 iter = m->private;
2467
2468 mutex_lock(&trace_types_lock);
2469 for_each_tracing_cpu(cpu) {
2470 if (iter->buffer_iter[cpu])
2471 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2472 }
2473
2474 if (iter->trace && iter->trace->close)
2475 iter->trace->close(iter);
2476
2477 /* reenable tracing if it was previously enabled */
2478 tracing_start();
2479 mutex_unlock(&trace_types_lock);
2480
2481 mutex_destroy(&iter->mutex);
2482 free_cpumask_var(iter->started);
2483 kfree(iter->trace);
2484 seq_release_private(inode, file);
2485 return 0;
2486}
2487
2488static int tracing_open(struct inode *inode, struct file *file)
2489{
2490 struct trace_iterator *iter;
2491 int ret = 0;
2492
2493 /* If this file was open for write, then erase contents */
2494 if ((file->f_mode & FMODE_WRITE) &&
2495 (file->f_flags & O_TRUNC)) {
2496 long cpu = (long) inode->i_private;
2497
2498 if (cpu == TRACE_PIPE_ALL_CPU)
2499 tracing_reset_online_cpus(&global_trace);
2500 else
2501 tracing_reset(&global_trace, cpu);
2502 }
2503
2504 if (file->f_mode & FMODE_READ) {
2505 iter = __tracing_open(inode, file);
2506 if (IS_ERR(iter))
2507 ret = PTR_ERR(iter);
2508 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2509 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2510 }
2511 return ret;
2512}
2513
2514static void *
2515t_next(struct seq_file *m, void *v, loff_t *pos)
2516{
2517 struct tracer *t = v;
2518
2519 (*pos)++;
2520
2521 if (t)
2522 t = t->next;
2523
2524 return t;
2525}
2526
2527static void *t_start(struct seq_file *m, loff_t *pos)
2528{
2529 struct tracer *t;
2530 loff_t l = 0;
2531
2532 mutex_lock(&trace_types_lock);
2533 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2534 ;
2535
2536 return t;
2537}
2538
2539static void t_stop(struct seq_file *m, void *p)
2540{
2541 mutex_unlock(&trace_types_lock);
2542}
2543
2544static int t_show(struct seq_file *m, void *v)
2545{
2546 struct tracer *t = v;
2547
2548 if (!t)
2549 return 0;
2550
2551 seq_printf(m, "%s", t->name);
2552 if (t->next)
2553 seq_putc(m, ' ');
2554 else
2555 seq_putc(m, '\n');
2556
2557 return 0;
2558}
2559
2560static const struct seq_operations show_traces_seq_ops = {
2561 .start = t_start,
2562 .next = t_next,
2563 .stop = t_stop,
2564 .show = t_show,
2565};
2566
2567static int show_traces_open(struct inode *inode, struct file *file)
2568{
2569 if (tracing_disabled)
2570 return -ENODEV;
2571
2572 return seq_open(file, &show_traces_seq_ops);
2573}
2574
2575static ssize_t
2576tracing_write_stub(struct file *filp, const char __user *ubuf,
2577 size_t count, loff_t *ppos)
2578{
2579 return count;
2580}
2581
2582static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2583{
2584 if (file->f_mode & FMODE_READ)
2585 return seq_lseek(file, offset, origin);
2586 else
2587 return 0;
2588}
2589
2590static const struct file_operations tracing_fops = {
2591 .open = tracing_open,
2592 .read = seq_read,
2593 .write = tracing_write_stub,
2594 .llseek = tracing_seek,
2595 .release = tracing_release,
2596};
2597
2598static const struct file_operations show_traces_fops = {
2599 .open = show_traces_open,
2600 .read = seq_read,
2601 .release = seq_release,
2602 .llseek = seq_lseek,
2603};
2604
2605/*
2606 * Only trace on a CPU if the bitmask is set:
2607 */
2608static cpumask_var_t tracing_cpumask;
2609
2610/*
2611 * The tracer itself will not take this lock, but still we want
2612 * to provide a consistent cpumask to user-space:
2613 */
2614static DEFINE_MUTEX(tracing_cpumask_update_lock);
2615
2616/*
2617 * Temporary storage for the character representation of the
2618 * CPU bitmask (and one more byte for the newline):
2619 */
2620static char mask_str[NR_CPUS + 1];
2621
2622static ssize_t
2623tracing_cpumask_read(struct file *filp, char __user *ubuf,
2624 size_t count, loff_t *ppos)
2625{
2626 int len;
2627
2628 mutex_lock(&tracing_cpumask_update_lock);
2629
2630 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2631 if (count - len < 2) {
2632 count = -EINVAL;
2633 goto out_err;
2634 }
2635 len += sprintf(mask_str + len, "\n");
2636 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2637
2638out_err:
2639 mutex_unlock(&tracing_cpumask_update_lock);
2640
2641 return count;
2642}
2643
2644static ssize_t
2645tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2646 size_t count, loff_t *ppos)
2647{
2648 int err, cpu;
2649 cpumask_var_t tracing_cpumask_new;
2650
2651 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2652 return -ENOMEM;
2653
2654 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2655 if (err)
2656 goto err_unlock;
2657
2658 mutex_lock(&tracing_cpumask_update_lock);
2659
2660 local_irq_disable();
2661 arch_spin_lock(&ftrace_max_lock);
2662 for_each_tracing_cpu(cpu) {
2663 /*
2664 * Increase/decrease the disabled counter if we are
2665 * about to flip a bit in the cpumask:
2666 */
2667 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2668 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2669 atomic_inc(&global_trace.data[cpu]->disabled);
2670 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2671 }
2672 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2673 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2674 atomic_dec(&global_trace.data[cpu]->disabled);
2675 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2676 }
2677 }
2678 arch_spin_unlock(&ftrace_max_lock);
2679 local_irq_enable();
2680
2681 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2682
2683 mutex_unlock(&tracing_cpumask_update_lock);
2684 free_cpumask_var(tracing_cpumask_new);
2685
2686 return count;
2687
2688err_unlock:
2689 free_cpumask_var(tracing_cpumask_new);
2690
2691 return err;
2692}
2693
2694static const struct file_operations tracing_cpumask_fops = {
2695 .open = tracing_open_generic,
2696 .read = tracing_cpumask_read,
2697 .write = tracing_cpumask_write,
2698 .llseek = generic_file_llseek,
2699};
2700
2701static int tracing_trace_options_show(struct seq_file *m, void *v)
2702{
2703 struct tracer_opt *trace_opts;
2704 u32 tracer_flags;
2705 int i;
2706
2707 mutex_lock(&trace_types_lock);
2708 tracer_flags = current_trace->flags->val;
2709 trace_opts = current_trace->flags->opts;
2710
2711 for (i = 0; trace_options[i]; i++) {
2712 if (trace_flags & (1 << i))
2713 seq_printf(m, "%s\n", trace_options[i]);
2714 else
2715 seq_printf(m, "no%s\n", trace_options[i]);
2716 }
2717
2718 for (i = 0; trace_opts[i].name; i++) {
2719 if (tracer_flags & trace_opts[i].bit)
2720 seq_printf(m, "%s\n", trace_opts[i].name);
2721 else
2722 seq_printf(m, "no%s\n", trace_opts[i].name);
2723 }
2724 mutex_unlock(&trace_types_lock);
2725
2726 return 0;
2727}
2728
2729static int __set_tracer_option(struct tracer *trace,
2730 struct tracer_flags *tracer_flags,
2731 struct tracer_opt *opts, int neg)
2732{
2733 int ret;
2734
2735 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2736 if (ret)
2737 return ret;
2738
2739 if (neg)
2740 tracer_flags->val &= ~opts->bit;
2741 else
2742 tracer_flags->val |= opts->bit;
2743 return 0;
2744}
2745
2746/* Try to assign a tracer specific option */
2747static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2748{
2749 struct tracer_flags *tracer_flags = trace->flags;
2750 struct tracer_opt *opts = NULL;
2751 int i;
2752
2753 for (i = 0; tracer_flags->opts[i].name; i++) {
2754 opts = &tracer_flags->opts[i];
2755
2756 if (strcmp(cmp, opts->name) == 0)
2757 return __set_tracer_option(trace, trace->flags,
2758 opts, neg);
2759 }
2760
2761 return -EINVAL;
2762}
2763
2764static void set_tracer_flags(unsigned int mask, int enabled)
2765{
2766 /* do nothing if flag is already set */
2767 if (!!(trace_flags & mask) == !!enabled)
2768 return;
2769
2770 if (enabled)
2771 trace_flags |= mask;
2772 else
2773 trace_flags &= ~mask;
2774
2775 if (mask == TRACE_ITER_RECORD_CMD)
2776 trace_event_enable_cmd_record(enabled);
2777
2778 if (mask == TRACE_ITER_OVERWRITE)
2779 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2780}
2781
2782static ssize_t
2783tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2784 size_t cnt, loff_t *ppos)
2785{
2786 char buf[64];
2787 char *cmp;
2788 int neg = 0;
2789 int ret;
2790 int i;
2791
2792 if (cnt >= sizeof(buf))
2793 return -EINVAL;
2794
2795 if (copy_from_user(&buf, ubuf, cnt))
2796 return -EFAULT;
2797
2798 buf[cnt] = 0;
2799 cmp = strstrip(buf);
2800
2801 if (strncmp(cmp, "no", 2) == 0) {
2802 neg = 1;
2803 cmp += 2;
2804 }
2805
2806 for (i = 0; trace_options[i]; i++) {
2807 if (strcmp(cmp, trace_options[i]) == 0) {
2808 set_tracer_flags(1 << i, !neg);
2809 break;
2810 }
2811 }
2812
2813 /* If no option could be set, test the specific tracer options */
2814 if (!trace_options[i]) {
2815 mutex_lock(&trace_types_lock);
2816 ret = set_tracer_option(current_trace, cmp, neg);
2817 mutex_unlock(&trace_types_lock);
2818 if (ret)
2819 return ret;
2820 }
2821
2822 *ppos += cnt;
2823
2824 return cnt;
2825}
2826
2827static int tracing_trace_options_open(struct inode *inode, struct file *file)
2828{
2829 if (tracing_disabled)
2830 return -ENODEV;
2831 return single_open(file, tracing_trace_options_show, NULL);
2832}
2833
2834static const struct file_operations tracing_iter_fops = {
2835 .open = tracing_trace_options_open,
2836 .read = seq_read,
2837 .llseek = seq_lseek,
2838 .release = single_release,
2839 .write = tracing_trace_options_write,
2840};
2841
2842static const char readme_msg[] =
2843 "tracing mini-HOWTO:\n\n"
2844 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2845 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2846 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2847 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2848 "nop\n"
2849 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2850 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2851 "wakeup\n"
2852 "# cat /sys/kernel/debug/tracing/trace_options\n"
2853 "noprint-parent nosym-offset nosym-addr noverbose\n"
2854 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2855 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2856 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2857 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2858;
2859
2860static ssize_t
2861tracing_readme_read(struct file *filp, char __user *ubuf,
2862 size_t cnt, loff_t *ppos)
2863{
2864 return simple_read_from_buffer(ubuf, cnt, ppos,
2865 readme_msg, strlen(readme_msg));
2866}
2867
2868static const struct file_operations tracing_readme_fops = {
2869 .open = tracing_open_generic,
2870 .read = tracing_readme_read,
2871 .llseek = generic_file_llseek,
2872};
2873
2874static ssize_t
2875tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2876 size_t cnt, loff_t *ppos)
2877{
2878 char *buf_comm;
2879 char *file_buf;
2880 char *buf;
2881 int len = 0;
2882 int pid;
2883 int i;
2884
2885 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2886 if (!file_buf)
2887 return -ENOMEM;
2888
2889 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2890 if (!buf_comm) {
2891 kfree(file_buf);
2892 return -ENOMEM;
2893 }
2894
2895 buf = file_buf;
2896
2897 for (i = 0; i < SAVED_CMDLINES; i++) {
2898 int r;
2899
2900 pid = map_cmdline_to_pid[i];
2901 if (pid == -1 || pid == NO_CMDLINE_MAP)
2902 continue;
2903
2904 trace_find_cmdline(pid, buf_comm);
2905 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2906 buf += r;
2907 len += r;
2908 }
2909
2910 len = simple_read_from_buffer(ubuf, cnt, ppos,
2911 file_buf, len);
2912
2913 kfree(file_buf);
2914 kfree(buf_comm);
2915
2916 return len;
2917}
2918
2919static const struct file_operations tracing_saved_cmdlines_fops = {
2920 .open = tracing_open_generic,
2921 .read = tracing_saved_cmdlines_read,
2922 .llseek = generic_file_llseek,
2923};
2924
2925static ssize_t
2926tracing_ctrl_read(struct file *filp, char __user *ubuf,
2927 size_t cnt, loff_t *ppos)
2928{
2929 char buf[64];
2930 int r;
2931
2932 r = sprintf(buf, "%u\n", tracer_enabled);
2933 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2934}
2935
2936static ssize_t
2937tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2938 size_t cnt, loff_t *ppos)
2939{
2940 struct trace_array *tr = filp->private_data;
2941 unsigned long val;
2942 int ret;
2943
2944 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2945 if (ret)
2946 return ret;
2947
2948 val = !!val;
2949
2950 mutex_lock(&trace_types_lock);
2951 if (tracer_enabled ^ val) {
2952
2953 /* Only need to warn if this is used to change the state */
2954 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2955
2956 if (val) {
2957 tracer_enabled = 1;
2958 if (current_trace->start)
2959 current_trace->start(tr);
2960 tracing_start();
2961 } else {
2962 tracer_enabled = 0;
2963 tracing_stop();
2964 if (current_trace->stop)
2965 current_trace->stop(tr);
2966 }
2967 }
2968 mutex_unlock(&trace_types_lock);
2969
2970 *ppos += cnt;
2971
2972 return cnt;
2973}
2974
2975static ssize_t
2976tracing_set_trace_read(struct file *filp, char __user *ubuf,
2977 size_t cnt, loff_t *ppos)
2978{
2979 char buf[MAX_TRACER_SIZE+2];
2980 int r;
2981
2982 mutex_lock(&trace_types_lock);
2983 if (current_trace)
2984 r = sprintf(buf, "%s\n", current_trace->name);
2985 else
2986 r = sprintf(buf, "\n");
2987 mutex_unlock(&trace_types_lock);
2988
2989 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2990}
2991
2992int tracer_init(struct tracer *t, struct trace_array *tr)
2993{
2994 tracing_reset_online_cpus(tr);
2995 return t->init(tr);
2996}
2997
2998static void set_buffer_entries(struct trace_array *tr, unsigned long val)
2999{
3000 int cpu;
3001 for_each_tracing_cpu(cpu)
3002 tr->data[cpu]->entries = val;
3003}
3004
3005static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3006{
3007 int ret;
3008
3009 /*
3010 * If kernel or user changes the size of the ring buffer
3011 * we use the size that was given, and we can forget about
3012 * expanding it later.
3013 */
3014 ring_buffer_expanded = 1;
3015
3016 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3017 if (ret < 0)
3018 return ret;
3019
3020 if (!current_trace->use_max_tr)
3021 goto out;
3022
3023 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3024 if (ret < 0) {
3025 int r = 0;
3026
3027 if (cpu == RING_BUFFER_ALL_CPUS) {
3028 int i;
3029 for_each_tracing_cpu(i) {
3030 r = ring_buffer_resize(global_trace.buffer,
3031 global_trace.data[i]->entries,
3032 i);
3033 if (r < 0)
3034 break;
3035 }
3036 } else {
3037 r = ring_buffer_resize(global_trace.buffer,
3038 global_trace.data[cpu]->entries,
3039 cpu);
3040 }
3041
3042 if (r < 0) {
3043 /*
3044 * AARGH! We are left with different
3045 * size max buffer!!!!
3046 * The max buffer is our "snapshot" buffer.
3047 * When a tracer needs a snapshot (one of the
3048 * latency tracers), it swaps the max buffer
3049 * with the saved snap shot. We succeeded to
3050 * update the size of the main buffer, but failed to
3051 * update the size of the max buffer. But when we tried
3052 * to reset the main buffer to the original size, we
3053 * failed there too. This is very unlikely to
3054 * happen, but if it does, warn and kill all
3055 * tracing.
3056 */
3057 WARN_ON(1);
3058 tracing_disabled = 1;
3059 }
3060 return ret;
3061 }
3062
3063 if (cpu == RING_BUFFER_ALL_CPUS)
3064 set_buffer_entries(&max_tr, size);
3065 else
3066 max_tr.data[cpu]->entries = size;
3067
3068 out:
3069 if (cpu == RING_BUFFER_ALL_CPUS)
3070 set_buffer_entries(&global_trace, size);
3071 else
3072 global_trace.data[cpu]->entries = size;
3073
3074 return ret;
3075}
3076
3077static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3078{
3079 int ret = size;
3080
3081 mutex_lock(&trace_types_lock);
3082
3083 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3084 /* make sure, this cpu is enabled in the mask */
3085 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3086 ret = -EINVAL;
3087 goto out;
3088 }
3089 }
3090
3091 ret = __tracing_resize_ring_buffer(size, cpu_id);
3092 if (ret < 0)
3093 ret = -ENOMEM;
3094
3095out:
3096 mutex_unlock(&trace_types_lock);
3097
3098 return ret;
3099}
3100
3101
3102/**
3103 * tracing_update_buffers - used by tracing facility to expand ring buffers
3104 *
3105 * To save on memory when the tracing is never used on a system with it
3106 * configured in. The ring buffers are set to a minimum size. But once
3107 * a user starts to use the tracing facility, then they need to grow
3108 * to their default size.
3109 *
3110 * This function is to be called when a tracer is about to be used.
3111 */
3112int tracing_update_buffers(void)
3113{
3114 int ret = 0;
3115
3116 mutex_lock(&trace_types_lock);
3117 if (!ring_buffer_expanded)
3118 ret = __tracing_resize_ring_buffer(trace_buf_size,
3119 RING_BUFFER_ALL_CPUS);
3120 mutex_unlock(&trace_types_lock);
3121
3122 return ret;
3123}
3124
3125struct trace_option_dentry;
3126
3127static struct trace_option_dentry *
3128create_trace_option_files(struct tracer *tracer);
3129
3130static void
3131destroy_trace_option_files(struct trace_option_dentry *topts);
3132
3133static int tracing_set_tracer(const char *buf)
3134{
3135 static struct trace_option_dentry *topts;
3136 struct trace_array *tr = &global_trace;
3137 struct tracer *t;
3138 int ret = 0;
3139
3140 mutex_lock(&trace_types_lock);
3141
3142 if (!ring_buffer_expanded) {
3143 ret = __tracing_resize_ring_buffer(trace_buf_size,
3144 RING_BUFFER_ALL_CPUS);
3145 if (ret < 0)
3146 goto out;
3147 ret = 0;
3148 }
3149
3150 for (t = trace_types; t; t = t->next) {
3151 if (strcmp(t->name, buf) == 0)
3152 break;
3153 }
3154 if (!t) {
3155 ret = -EINVAL;
3156 goto out;
3157 }
3158 if (t == current_trace)
3159 goto out;
3160
3161 trace_branch_disable();
3162 if (current_trace && current_trace->reset)
3163 current_trace->reset(tr);
3164 if (current_trace && current_trace->use_max_tr) {
3165 /*
3166 * We don't free the ring buffer. instead, resize it because
3167 * The max_tr ring buffer has some state (e.g. ring->clock) and
3168 * we want preserve it.
3169 */
3170 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3171 set_buffer_entries(&max_tr, 1);
3172 }
3173 destroy_trace_option_files(topts);
3174
3175 current_trace = t;
3176
3177 topts = create_trace_option_files(current_trace);
3178 if (current_trace->use_max_tr) {
3179 int cpu;
3180 /* we need to make per cpu buffer sizes equivalent */
3181 for_each_tracing_cpu(cpu) {
3182 ret = ring_buffer_resize(max_tr.buffer,
3183 global_trace.data[cpu]->entries,
3184 cpu);
3185 if (ret < 0)
3186 goto out;
3187 max_tr.data[cpu]->entries =
3188 global_trace.data[cpu]->entries;
3189 }
3190 }
3191
3192 if (t->init) {
3193 ret = tracer_init(t, tr);
3194 if (ret)
3195 goto out;
3196 }
3197
3198 trace_branch_enable(tr);
3199 out:
3200 mutex_unlock(&trace_types_lock);
3201
3202 return ret;
3203}
3204
3205static ssize_t
3206tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3207 size_t cnt, loff_t *ppos)
3208{
3209 char buf[MAX_TRACER_SIZE+1];
3210 int i;
3211 size_t ret;
3212 int err;
3213
3214 ret = cnt;
3215
3216 if (cnt > MAX_TRACER_SIZE)
3217 cnt = MAX_TRACER_SIZE;
3218
3219 if (copy_from_user(&buf, ubuf, cnt))
3220 return -EFAULT;
3221
3222 buf[cnt] = 0;
3223
3224 /* strip ending whitespace. */
3225 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3226 buf[i] = 0;
3227
3228 err = tracing_set_tracer(buf);
3229 if (err)
3230 return err;
3231
3232 *ppos += ret;
3233
3234 return ret;
3235}
3236
3237static ssize_t
3238tracing_max_lat_read(struct file *filp, char __user *ubuf,
3239 size_t cnt, loff_t *ppos)
3240{
3241 unsigned long *ptr = filp->private_data;
3242 char buf[64];
3243 int r;
3244
3245 r = snprintf(buf, sizeof(buf), "%ld\n",
3246 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3247 if (r > sizeof(buf))
3248 r = sizeof(buf);
3249 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3250}
3251
3252static ssize_t
3253tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3254 size_t cnt, loff_t *ppos)
3255{
3256 unsigned long *ptr = filp->private_data;
3257 unsigned long val;
3258 int ret;
3259
3260 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3261 if (ret)
3262 return ret;
3263
3264 *ptr = val * 1000;
3265
3266 return cnt;
3267}
3268
3269static int tracing_open_pipe(struct inode *inode, struct file *filp)
3270{
3271 long cpu_file = (long) inode->i_private;
3272 struct trace_iterator *iter;
3273 int ret = 0;
3274
3275 if (tracing_disabled)
3276 return -ENODEV;
3277
3278 mutex_lock(&trace_types_lock);
3279
3280 /* create a buffer to store the information to pass to userspace */
3281 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3282 if (!iter) {
3283 ret = -ENOMEM;
3284 goto out;
3285 }
3286
3287 /*
3288 * We make a copy of the current tracer to avoid concurrent
3289 * changes on it while we are reading.
3290 */
3291 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3292 if (!iter->trace) {
3293 ret = -ENOMEM;
3294 goto fail;
3295 }
3296 if (current_trace)
3297 *iter->trace = *current_trace;
3298
3299 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3300 ret = -ENOMEM;
3301 goto fail;
3302 }
3303
3304 /* trace pipe does not show start of buffer */
3305 cpumask_setall(iter->started);
3306
3307 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3308 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3309
3310 iter->cpu_file = cpu_file;
3311 iter->tr = &global_trace;
3312 mutex_init(&iter->mutex);
3313 filp->private_data = iter;
3314
3315 if (iter->trace->pipe_open)
3316 iter->trace->pipe_open(iter);
3317
3318 nonseekable_open(inode, filp);
3319out:
3320 mutex_unlock(&trace_types_lock);
3321 return ret;
3322
3323fail:
3324 kfree(iter->trace);
3325 kfree(iter);
3326 mutex_unlock(&trace_types_lock);
3327 return ret;
3328}
3329
3330static int tracing_release_pipe(struct inode *inode, struct file *file)
3331{
3332 struct trace_iterator *iter = file->private_data;
3333
3334 mutex_lock(&trace_types_lock);
3335
3336 if (iter->trace->pipe_close)
3337 iter->trace->pipe_close(iter);
3338
3339 mutex_unlock(&trace_types_lock);
3340
3341 free_cpumask_var(iter->started);
3342 mutex_destroy(&iter->mutex);
3343 kfree(iter->trace);
3344 kfree(iter);
3345
3346 return 0;
3347}
3348
3349static unsigned int
3350tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3351{
3352 struct trace_iterator *iter = filp->private_data;
3353
3354 if (trace_flags & TRACE_ITER_BLOCK) {
3355 /*
3356 * Always select as readable when in blocking mode
3357 */
3358 return POLLIN | POLLRDNORM;
3359 } else {
3360 if (!trace_empty(iter))
3361 return POLLIN | POLLRDNORM;
3362 poll_wait(filp, &trace_wait, poll_table);
3363 if (!trace_empty(iter))
3364 return POLLIN | POLLRDNORM;
3365
3366 return 0;
3367 }
3368}
3369
3370
3371void default_wait_pipe(struct trace_iterator *iter)
3372{
3373 DEFINE_WAIT(wait);
3374
3375 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3376
3377 if (trace_empty(iter))
3378 schedule();
3379
3380 finish_wait(&trace_wait, &wait);
3381}
3382
3383/*
3384 * This is a make-shift waitqueue.
3385 * A tracer might use this callback on some rare cases:
3386 *
3387 * 1) the current tracer might hold the runqueue lock when it wakes up
3388 * a reader, hence a deadlock (sched, function, and function graph tracers)
3389 * 2) the function tracers, trace all functions, we don't want
3390 * the overhead of calling wake_up and friends
3391 * (and tracing them too)
3392 *
3393 * Anyway, this is really very primitive wakeup.
3394 */
3395void poll_wait_pipe(struct trace_iterator *iter)
3396{
3397 set_current_state(TASK_INTERRUPTIBLE);
3398 /* sleep for 100 msecs, and try again. */
3399 schedule_timeout(HZ / 10);
3400}
3401
3402/* Must be called with trace_types_lock mutex held. */
3403static int tracing_wait_pipe(struct file *filp)
3404{
3405 struct trace_iterator *iter = filp->private_data;
3406
3407 while (trace_empty(iter)) {
3408
3409 if ((filp->f_flags & O_NONBLOCK)) {
3410 return -EAGAIN;
3411 }
3412
3413 mutex_unlock(&iter->mutex);
3414
3415 iter->trace->wait_pipe(iter);
3416
3417 mutex_lock(&iter->mutex);
3418
3419 if (signal_pending(current))
3420 return -EINTR;
3421
3422 /*
3423 * We block until we read something and tracing is disabled.
3424 * We still block if tracing is disabled, but we have never
3425 * read anything. This allows a user to cat this file, and
3426 * then enable tracing. But after we have read something,
3427 * we give an EOF when tracing is again disabled.
3428 *
3429 * iter->pos will be 0 if we haven't read anything.
3430 */
3431 if (!tracer_enabled && iter->pos)
3432 break;
3433 }
3434
3435 return 1;
3436}
3437
3438/*
3439 * Consumer reader.
3440 */
3441static ssize_t
3442tracing_read_pipe(struct file *filp, char __user *ubuf,
3443 size_t cnt, loff_t *ppos)
3444{
3445 struct trace_iterator *iter = filp->private_data;
3446 static struct tracer *old_tracer;
3447 ssize_t sret;
3448
3449 /* return any leftover data */
3450 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3451 if (sret != -EBUSY)
3452 return sret;
3453
3454 trace_seq_init(&iter->seq);
3455
3456 /* copy the tracer to avoid using a global lock all around */
3457 mutex_lock(&trace_types_lock);
3458 if (unlikely(old_tracer != current_trace && current_trace)) {
3459 old_tracer = current_trace;
3460 *iter->trace = *current_trace;
3461 }
3462 mutex_unlock(&trace_types_lock);
3463
3464 /*
3465 * Avoid more than one consumer on a single file descriptor
3466 * This is just a matter of traces coherency, the ring buffer itself
3467 * is protected.
3468 */
3469 mutex_lock(&iter->mutex);
3470 if (iter->trace->read) {
3471 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3472 if (sret)
3473 goto out;
3474 }
3475
3476waitagain:
3477 sret = tracing_wait_pipe(filp);
3478 if (sret <= 0)
3479 goto out;
3480
3481 /* stop when tracing is finished */
3482 if (trace_empty(iter)) {
3483 sret = 0;
3484 goto out;
3485 }
3486
3487 if (cnt >= PAGE_SIZE)
3488 cnt = PAGE_SIZE - 1;
3489
3490 /* reset all but tr, trace, and overruns */
3491 memset(&iter->seq, 0,
3492 sizeof(struct trace_iterator) -
3493 offsetof(struct trace_iterator, seq));
3494 iter->pos = -1;
3495
3496 trace_event_read_lock();
3497 trace_access_lock(iter->cpu_file);
3498 while (trace_find_next_entry_inc(iter) != NULL) {
3499 enum print_line_t ret;
3500 int len = iter->seq.len;
3501
3502 ret = print_trace_line(iter);
3503 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3504 /* don't print partial lines */
3505 iter->seq.len = len;
3506 break;
3507 }
3508 if (ret != TRACE_TYPE_NO_CONSUME)
3509 trace_consume(iter);
3510
3511 if (iter->seq.len >= cnt)
3512 break;
3513
3514 /*
3515 * Setting the full flag means we reached the trace_seq buffer
3516 * size and we should leave by partial output condition above.
3517 * One of the trace_seq_* functions is not used properly.
3518 */
3519 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3520 iter->ent->type);
3521 }
3522 trace_access_unlock(iter->cpu_file);
3523 trace_event_read_unlock();
3524
3525 /* Now copy what we have to the user */
3526 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3527 if (iter->seq.readpos >= iter->seq.len)
3528 trace_seq_init(&iter->seq);
3529
3530 /*
3531 * If there was nothing to send to user, in spite of consuming trace
3532 * entries, go back to wait for more entries.
3533 */
3534 if (sret == -EBUSY)
3535 goto waitagain;
3536
3537out:
3538 mutex_unlock(&iter->mutex);
3539
3540 return sret;
3541}
3542
3543static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3544 struct pipe_buffer *buf)
3545{
3546 __free_page(buf->page);
3547}
3548
3549static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3550 unsigned int idx)
3551{
3552 __free_page(spd->pages[idx]);
3553}
3554
3555static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3556 .can_merge = 0,
3557 .map = generic_pipe_buf_map,
3558 .unmap = generic_pipe_buf_unmap,
3559 .confirm = generic_pipe_buf_confirm,
3560 .release = tracing_pipe_buf_release,
3561 .steal = generic_pipe_buf_steal,
3562 .get = generic_pipe_buf_get,
3563};
3564
3565static size_t
3566tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3567{
3568 size_t count;
3569 int ret;
3570
3571 /* Seq buffer is page-sized, exactly what we need. */
3572 for (;;) {
3573 count = iter->seq.len;
3574 ret = print_trace_line(iter);
3575 count = iter->seq.len - count;
3576 if (rem < count) {
3577 rem = 0;
3578 iter->seq.len -= count;
3579 break;
3580 }
3581 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3582 iter->seq.len -= count;
3583 break;
3584 }
3585
3586 if (ret != TRACE_TYPE_NO_CONSUME)
3587 trace_consume(iter);
3588 rem -= count;
3589 if (!trace_find_next_entry_inc(iter)) {
3590 rem = 0;
3591 iter->ent = NULL;
3592 break;
3593 }
3594 }
3595
3596 return rem;
3597}
3598
3599static ssize_t tracing_splice_read_pipe(struct file *filp,
3600 loff_t *ppos,
3601 struct pipe_inode_info *pipe,
3602 size_t len,
3603 unsigned int flags)
3604{
3605 struct page *pages_def[PIPE_DEF_BUFFERS];
3606 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3607 struct trace_iterator *iter = filp->private_data;
3608 struct splice_pipe_desc spd = {
3609 .pages = pages_def,
3610 .partial = partial_def,
3611 .nr_pages = 0, /* This gets updated below. */
3612 .nr_pages_max = PIPE_DEF_BUFFERS,
3613 .flags = flags,
3614 .ops = &tracing_pipe_buf_ops,
3615 .spd_release = tracing_spd_release_pipe,
3616 };
3617 static struct tracer *old_tracer;
3618 ssize_t ret;
3619 size_t rem;
3620 unsigned int i;
3621
3622 if (splice_grow_spd(pipe, &spd))
3623 return -ENOMEM;
3624
3625 /* copy the tracer to avoid using a global lock all around */
3626 mutex_lock(&trace_types_lock);
3627 if (unlikely(old_tracer != current_trace && current_trace)) {
3628 old_tracer = current_trace;
3629 *iter->trace = *current_trace;
3630 }
3631 mutex_unlock(&trace_types_lock);
3632
3633 mutex_lock(&iter->mutex);
3634
3635 if (iter->trace->splice_read) {
3636 ret = iter->trace->splice_read(iter, filp,
3637 ppos, pipe, len, flags);
3638 if (ret)
3639 goto out_err;
3640 }
3641
3642 ret = tracing_wait_pipe(filp);
3643 if (ret <= 0)
3644 goto out_err;
3645
3646 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3647 ret = -EFAULT;
3648 goto out_err;
3649 }
3650
3651 trace_event_read_lock();
3652 trace_access_lock(iter->cpu_file);
3653
3654 /* Fill as many pages as possible. */
3655 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3656 spd.pages[i] = alloc_page(GFP_KERNEL);
3657 if (!spd.pages[i])
3658 break;
3659
3660 rem = tracing_fill_pipe_page(rem, iter);
3661
3662 /* Copy the data into the page, so we can start over. */
3663 ret = trace_seq_to_buffer(&iter->seq,
3664 page_address(spd.pages[i]),
3665 iter->seq.len);
3666 if (ret < 0) {
3667 __free_page(spd.pages[i]);
3668 break;
3669 }
3670 spd.partial[i].offset = 0;
3671 spd.partial[i].len = iter->seq.len;
3672
3673 trace_seq_init(&iter->seq);
3674 }
3675
3676 trace_access_unlock(iter->cpu_file);
3677 trace_event_read_unlock();
3678 mutex_unlock(&iter->mutex);
3679
3680 spd.nr_pages = i;
3681
3682 ret = splice_to_pipe(pipe, &spd);
3683out:
3684 splice_shrink_spd(&spd);
3685 return ret;
3686
3687out_err:
3688 mutex_unlock(&iter->mutex);
3689 goto out;
3690}
3691
3692struct ftrace_entries_info {
3693 struct trace_array *tr;
3694 int cpu;
3695};
3696
3697static int tracing_entries_open(struct inode *inode, struct file *filp)
3698{
3699 struct ftrace_entries_info *info;
3700
3701 if (tracing_disabled)
3702 return -ENODEV;
3703
3704 info = kzalloc(sizeof(*info), GFP_KERNEL);
3705 if (!info)
3706 return -ENOMEM;
3707
3708 info->tr = &global_trace;
3709 info->cpu = (unsigned long)inode->i_private;
3710
3711 filp->private_data = info;
3712
3713 return 0;
3714}
3715
3716static ssize_t
3717tracing_entries_read(struct file *filp, char __user *ubuf,
3718 size_t cnt, loff_t *ppos)
3719{
3720 struct ftrace_entries_info *info = filp->private_data;
3721 struct trace_array *tr = info->tr;
3722 char buf[64];
3723 int r = 0;
3724 ssize_t ret;
3725
3726 mutex_lock(&trace_types_lock);
3727
3728 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3729 int cpu, buf_size_same;
3730 unsigned long size;
3731
3732 size = 0;
3733 buf_size_same = 1;
3734 /* check if all cpu sizes are same */
3735 for_each_tracing_cpu(cpu) {
3736 /* fill in the size from first enabled cpu */
3737 if (size == 0)
3738 size = tr->data[cpu]->entries;
3739 if (size != tr->data[cpu]->entries) {
3740 buf_size_same = 0;
3741 break;
3742 }
3743 }
3744
3745 if (buf_size_same) {
3746 if (!ring_buffer_expanded)
3747 r = sprintf(buf, "%lu (expanded: %lu)\n",
3748 size >> 10,
3749 trace_buf_size >> 10);
3750 else
3751 r = sprintf(buf, "%lu\n", size >> 10);
3752 } else
3753 r = sprintf(buf, "X\n");
3754 } else
3755 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3756
3757 mutex_unlock(&trace_types_lock);
3758
3759 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3760 return ret;
3761}
3762
3763static ssize_t
3764tracing_entries_write(struct file *filp, const char __user *ubuf,
3765 size_t cnt, loff_t *ppos)
3766{
3767 struct ftrace_entries_info *info = filp->private_data;
3768 unsigned long val;
3769 int ret;
3770
3771 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3772 if (ret)
3773 return ret;
3774
3775 /* must have at least 1 entry */
3776 if (!val)
3777 return -EINVAL;
3778
3779 /* value is in KB */
3780 val <<= 10;
3781
3782 ret = tracing_resize_ring_buffer(val, info->cpu);
3783 if (ret < 0)
3784 return ret;
3785
3786 *ppos += cnt;
3787
3788 return cnt;
3789}
3790
3791static int
3792tracing_entries_release(struct inode *inode, struct file *filp)
3793{
3794 struct ftrace_entries_info *info = filp->private_data;
3795
3796 kfree(info);
3797
3798 return 0;
3799}
3800
3801static ssize_t
3802tracing_total_entries_read(struct file *filp, char __user *ubuf,
3803 size_t cnt, loff_t *ppos)
3804{
3805 struct trace_array *tr = filp->private_data;
3806 char buf[64];
3807 int r, cpu;
3808 unsigned long size = 0, expanded_size = 0;
3809
3810 mutex_lock(&trace_types_lock);
3811 for_each_tracing_cpu(cpu) {
3812 size += tr->data[cpu]->entries >> 10;
3813 if (!ring_buffer_expanded)
3814 expanded_size += trace_buf_size >> 10;
3815 }
3816 if (ring_buffer_expanded)
3817 r = sprintf(buf, "%lu\n", size);
3818 else
3819 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3820 mutex_unlock(&trace_types_lock);
3821
3822 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3823}
3824
3825static ssize_t
3826tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3827 size_t cnt, loff_t *ppos)
3828{
3829 /*
3830 * There is no need to read what the user has written, this function
3831 * is just to make sure that there is no error when "echo" is used
3832 */
3833
3834 *ppos += cnt;
3835
3836 return cnt;
3837}
3838
3839static int
3840tracing_free_buffer_release(struct inode *inode, struct file *filp)
3841{
3842 /* disable tracing ? */
3843 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3844 tracing_off();
3845 /* resize the ring buffer to 0 */
3846 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3847
3848 return 0;
3849}
3850
3851static ssize_t
3852tracing_mark_write(struct file *filp, const char __user *ubuf,
3853 size_t cnt, loff_t *fpos)
3854{
3855 unsigned long addr = (unsigned long)ubuf;
3856 struct ring_buffer_event *event;
3857 struct ring_buffer *buffer;
3858 struct print_entry *entry;
3859 unsigned long irq_flags;
3860 struct page *pages[2];
3861 void *map_page[2];
3862 int nr_pages = 1;
3863 ssize_t written;
3864 int offset;
3865 int size;
3866 int len;
3867 int ret;
3868 int i;
3869
3870 if (tracing_disabled)
3871 return -EINVAL;
3872
3873 if (cnt > TRACE_BUF_SIZE)
3874 cnt = TRACE_BUF_SIZE;
3875
3876 /*
3877 * Userspace is injecting traces into the kernel trace buffer.
3878 * We want to be as non intrusive as possible.
3879 * To do so, we do not want to allocate any special buffers
3880 * or take any locks, but instead write the userspace data
3881 * straight into the ring buffer.
3882 *
3883 * First we need to pin the userspace buffer into memory,
3884 * which, most likely it is, because it just referenced it.
3885 * But there's no guarantee that it is. By using get_user_pages_fast()
3886 * and kmap_atomic/kunmap_atomic() we can get access to the
3887 * pages directly. We then write the data directly into the
3888 * ring buffer.
3889 */
3890 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3891
3892 /* check if we cross pages */
3893 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3894 nr_pages = 2;
3895
3896 offset = addr & (PAGE_SIZE - 1);
3897 addr &= PAGE_MASK;
3898
3899 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3900 if (ret < nr_pages) {
3901 while (--ret >= 0)
3902 put_page(pages[ret]);
3903 written = -EFAULT;
3904 goto out;
3905 }
3906
3907 for (i = 0; i < nr_pages; i++)
3908 map_page[i] = kmap_atomic(pages[i]);
3909
3910 local_save_flags(irq_flags);
3911 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3912 buffer = global_trace.buffer;
3913 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3914 irq_flags, preempt_count());
3915 if (!event) {
3916 /* Ring buffer disabled, return as if not open for write */
3917 written = -EBADF;
3918 goto out_unlock;
3919 }
3920
3921 entry = ring_buffer_event_data(event);
3922 entry->ip = _THIS_IP_;
3923
3924 if (nr_pages == 2) {
3925 len = PAGE_SIZE - offset;
3926 memcpy(&entry->buf, map_page[0] + offset, len);
3927 memcpy(&entry->buf[len], map_page[1], cnt - len);
3928 } else
3929 memcpy(&entry->buf, map_page[0] + offset, cnt);
3930
3931 if (entry->buf[cnt - 1] != '\n') {
3932 entry->buf[cnt] = '\n';
3933 entry->buf[cnt + 1] = '\0';
3934 } else
3935 entry->buf[cnt] = '\0';
3936
3937 ring_buffer_unlock_commit(buffer, event);
3938
3939 written = cnt;
3940
3941 *fpos += written;
3942
3943 out_unlock:
3944 for (i = 0; i < nr_pages; i++){
3945 kunmap_atomic(map_page[i]);
3946 put_page(pages[i]);
3947 }
3948 out:
3949 return written;
3950}
3951
3952static int tracing_clock_show(struct seq_file *m, void *v)
3953{
3954 int i;
3955
3956 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3957 seq_printf(m,
3958 "%s%s%s%s", i ? " " : "",
3959 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3960 i == trace_clock_id ? "]" : "");
3961 seq_putc(m, '\n');
3962
3963 return 0;
3964}
3965
3966static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3967 size_t cnt, loff_t *fpos)
3968{
3969 char buf[64];
3970 const char *clockstr;
3971 int i;
3972
3973 if (cnt >= sizeof(buf))
3974 return -EINVAL;
3975
3976 if (copy_from_user(&buf, ubuf, cnt))
3977 return -EFAULT;
3978
3979 buf[cnt] = 0;
3980
3981 clockstr = strstrip(buf);
3982
3983 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3984 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3985 break;
3986 }
3987 if (i == ARRAY_SIZE(trace_clocks))
3988 return -EINVAL;
3989
3990 trace_clock_id = i;
3991
3992 mutex_lock(&trace_types_lock);
3993
3994 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3995 if (max_tr.buffer)
3996 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3997
3998 mutex_unlock(&trace_types_lock);
3999
4000 *fpos += cnt;
4001
4002 return cnt;
4003}
4004
4005static int tracing_clock_open(struct inode *inode, struct file *file)
4006{
4007 if (tracing_disabled)
4008 return -ENODEV;
4009 return single_open(file, tracing_clock_show, NULL);
4010}
4011
4012static const struct file_operations tracing_max_lat_fops = {
4013 .open = tracing_open_generic,
4014 .read = tracing_max_lat_read,
4015 .write = tracing_max_lat_write,
4016 .llseek = generic_file_llseek,
4017};
4018
4019static const struct file_operations tracing_ctrl_fops = {
4020 .open = tracing_open_generic,
4021 .read = tracing_ctrl_read,
4022 .write = tracing_ctrl_write,
4023 .llseek = generic_file_llseek,
4024};
4025
4026static const struct file_operations set_tracer_fops = {
4027 .open = tracing_open_generic,
4028 .read = tracing_set_trace_read,
4029 .write = tracing_set_trace_write,
4030 .llseek = generic_file_llseek,
4031};
4032
4033static const struct file_operations tracing_pipe_fops = {
4034 .open = tracing_open_pipe,
4035 .poll = tracing_poll_pipe,
4036 .read = tracing_read_pipe,
4037 .splice_read = tracing_splice_read_pipe,
4038 .release = tracing_release_pipe,
4039 .llseek = no_llseek,
4040};
4041
4042static const struct file_operations tracing_entries_fops = {
4043 .open = tracing_entries_open,
4044 .read = tracing_entries_read,
4045 .write = tracing_entries_write,
4046 .release = tracing_entries_release,
4047 .llseek = generic_file_llseek,
4048};
4049
4050static const struct file_operations tracing_total_entries_fops = {
4051 .open = tracing_open_generic,
4052 .read = tracing_total_entries_read,
4053 .llseek = generic_file_llseek,
4054};
4055
4056static const struct file_operations tracing_free_buffer_fops = {
4057 .write = tracing_free_buffer_write,
4058 .release = tracing_free_buffer_release,
4059};
4060
4061static const struct file_operations tracing_mark_fops = {
4062 .open = tracing_open_generic,
4063 .write = tracing_mark_write,
4064 .llseek = generic_file_llseek,
4065};
4066
4067static const struct file_operations trace_clock_fops = {
4068 .open = tracing_clock_open,
4069 .read = seq_read,
4070 .llseek = seq_lseek,
4071 .release = single_release,
4072 .write = tracing_clock_write,
4073};
4074
4075struct ftrace_buffer_info {
4076 struct trace_array *tr;
4077 void *spare;
4078 int cpu;
4079 unsigned int read;
4080};
4081
4082static int tracing_buffers_open(struct inode *inode, struct file *filp)
4083{
4084 int cpu = (int)(long)inode->i_private;
4085 struct ftrace_buffer_info *info;
4086
4087 if (tracing_disabled)
4088 return -ENODEV;
4089
4090 info = kzalloc(sizeof(*info), GFP_KERNEL);
4091 if (!info)
4092 return -ENOMEM;
4093
4094 info->tr = &global_trace;
4095 info->cpu = cpu;
4096 info->spare = NULL;
4097 /* Force reading ring buffer for first read */
4098 info->read = (unsigned int)-1;
4099
4100 filp->private_data = info;
4101
4102 return nonseekable_open(inode, filp);
4103}
4104
4105static ssize_t
4106tracing_buffers_read(struct file *filp, char __user *ubuf,
4107 size_t count, loff_t *ppos)
4108{
4109 struct ftrace_buffer_info *info = filp->private_data;
4110 ssize_t ret;
4111 size_t size;
4112
4113 if (!count)
4114 return 0;
4115
4116 if (!info->spare)
4117 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4118 if (!info->spare)
4119 return -ENOMEM;
4120
4121 /* Do we have previous read data to read? */
4122 if (info->read < PAGE_SIZE)
4123 goto read;
4124
4125 trace_access_lock(info->cpu);
4126 ret = ring_buffer_read_page(info->tr->buffer,
4127 &info->spare,
4128 count,
4129 info->cpu, 0);
4130 trace_access_unlock(info->cpu);
4131 if (ret < 0)
4132 return 0;
4133
4134 info->read = 0;
4135
4136read:
4137 size = PAGE_SIZE - info->read;
4138 if (size > count)
4139 size = count;
4140
4141 ret = copy_to_user(ubuf, info->spare + info->read, size);
4142 if (ret == size)
4143 return -EFAULT;
4144 size -= ret;
4145
4146 *ppos += size;
4147 info->read += size;
4148
4149 return size;
4150}
4151
4152static int tracing_buffers_release(struct inode *inode, struct file *file)
4153{
4154 struct ftrace_buffer_info *info = file->private_data;
4155
4156 if (info->spare)
4157 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4158 kfree(info);
4159
4160 return 0;
4161}
4162
4163struct buffer_ref {
4164 struct ring_buffer *buffer;
4165 void *page;
4166 int ref;
4167};
4168
4169static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4170 struct pipe_buffer *buf)
4171{
4172 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4173
4174 if (--ref->ref)
4175 return;
4176
4177 ring_buffer_free_read_page(ref->buffer, ref->page);
4178 kfree(ref);
4179 buf->private = 0;
4180}
4181
4182static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4183 struct pipe_buffer *buf)
4184{
4185 return 1;
4186}
4187
4188static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4189 struct pipe_buffer *buf)
4190{
4191 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4192
4193 ref->ref++;
4194}
4195
4196/* Pipe buffer operations for a buffer. */
4197static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4198 .can_merge = 0,
4199 .map = generic_pipe_buf_map,
4200 .unmap = generic_pipe_buf_unmap,
4201 .confirm = generic_pipe_buf_confirm,
4202 .release = buffer_pipe_buf_release,
4203 .steal = buffer_pipe_buf_steal,
4204 .get = buffer_pipe_buf_get,
4205};
4206
4207/*
4208 * Callback from splice_to_pipe(), if we need to release some pages
4209 * at the end of the spd in case we error'ed out in filling the pipe.
4210 */
4211static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4212{
4213 struct buffer_ref *ref =
4214 (struct buffer_ref *)spd->partial[i].private;
4215
4216 if (--ref->ref)
4217 return;
4218
4219 ring_buffer_free_read_page(ref->buffer, ref->page);
4220 kfree(ref);
4221 spd->partial[i].private = 0;
4222}
4223
4224static ssize_t
4225tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4226 struct pipe_inode_info *pipe, size_t len,
4227 unsigned int flags)
4228{
4229 struct ftrace_buffer_info *info = file->private_data;
4230 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4231 struct page *pages_def[PIPE_DEF_BUFFERS];
4232 struct splice_pipe_desc spd = {
4233 .pages = pages_def,
4234 .partial = partial_def,
4235 .nr_pages_max = PIPE_DEF_BUFFERS,
4236 .flags = flags,
4237 .ops = &buffer_pipe_buf_ops,
4238 .spd_release = buffer_spd_release,
4239 };
4240 struct buffer_ref *ref;
4241 int entries, size, i;
4242 size_t ret;
4243
4244 if (splice_grow_spd(pipe, &spd))
4245 return -ENOMEM;
4246
4247 if (*ppos & (PAGE_SIZE - 1)) {
4248 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4249 ret = -EINVAL;
4250 goto out;
4251 }
4252
4253 if (len & (PAGE_SIZE - 1)) {
4254 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4255 if (len < PAGE_SIZE) {
4256 ret = -EINVAL;
4257 goto out;
4258 }
4259 len &= PAGE_MASK;
4260 }
4261
4262 trace_access_lock(info->cpu);
4263 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4264
4265 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4266 struct page *page;
4267 int r;
4268
4269 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4270 if (!ref)
4271 break;
4272
4273 ref->ref = 1;
4274 ref->buffer = info->tr->buffer;
4275 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4276 if (!ref->page) {
4277 kfree(ref);
4278 break;
4279 }
4280
4281 r = ring_buffer_read_page(ref->buffer, &ref->page,
4282 len, info->cpu, 1);
4283 if (r < 0) {
4284 ring_buffer_free_read_page(ref->buffer, ref->page);
4285 kfree(ref);
4286 break;
4287 }
4288
4289 /*
4290 * zero out any left over data, this is going to
4291 * user land.
4292 */
4293 size = ring_buffer_page_len(ref->page);
4294 if (size < PAGE_SIZE)
4295 memset(ref->page + size, 0, PAGE_SIZE - size);
4296
4297 page = virt_to_page(ref->page);
4298
4299 spd.pages[i] = page;
4300 spd.partial[i].len = PAGE_SIZE;
4301 spd.partial[i].offset = 0;
4302 spd.partial[i].private = (unsigned long)ref;
4303 spd.nr_pages++;
4304 *ppos += PAGE_SIZE;
4305
4306 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4307 }
4308
4309 trace_access_unlock(info->cpu);
4310 spd.nr_pages = i;
4311
4312 /* did we read anything? */
4313 if (!spd.nr_pages) {
4314 if (flags & SPLICE_F_NONBLOCK)
4315 ret = -EAGAIN;
4316 else
4317 ret = 0;
4318 /* TODO: block */
4319 goto out;
4320 }
4321
4322 ret = splice_to_pipe(pipe, &spd);
4323 splice_shrink_spd(&spd);
4324out:
4325 return ret;
4326}
4327
4328static const struct file_operations tracing_buffers_fops = {
4329 .open = tracing_buffers_open,
4330 .read = tracing_buffers_read,
4331 .release = tracing_buffers_release,
4332 .splice_read = tracing_buffers_splice_read,
4333 .llseek = no_llseek,
4334};
4335
4336static ssize_t
4337tracing_stats_read(struct file *filp, char __user *ubuf,
4338 size_t count, loff_t *ppos)
4339{
4340 unsigned long cpu = (unsigned long)filp->private_data;
4341 struct trace_array *tr = &global_trace;
4342 struct trace_seq *s;
4343 unsigned long cnt;
4344 unsigned long long t;
4345 unsigned long usec_rem;
4346
4347 s = kmalloc(sizeof(*s), GFP_KERNEL);
4348 if (!s)
4349 return -ENOMEM;
4350
4351 trace_seq_init(s);
4352
4353 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4354 trace_seq_printf(s, "entries: %ld\n", cnt);
4355
4356 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4357 trace_seq_printf(s, "overrun: %ld\n", cnt);
4358
4359 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4360 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4361
4362 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4363 trace_seq_printf(s, "bytes: %ld\n", cnt);
4364
4365 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4366 usec_rem = do_div(t, USEC_PER_SEC);
4367 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4368
4369 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4370 usec_rem = do_div(t, USEC_PER_SEC);
4371 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4372
4373 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4374
4375 kfree(s);
4376
4377 return count;
4378}
4379
4380static const struct file_operations tracing_stats_fops = {
4381 .open = tracing_open_generic,
4382 .read = tracing_stats_read,
4383 .llseek = generic_file_llseek,
4384};
4385
4386#ifdef CONFIG_DYNAMIC_FTRACE
4387
4388int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4389{
4390 return 0;
4391}
4392
4393static ssize_t
4394tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4395 size_t cnt, loff_t *ppos)
4396{
4397 static char ftrace_dyn_info_buffer[1024];
4398 static DEFINE_MUTEX(dyn_info_mutex);
4399 unsigned long *p = filp->private_data;
4400 char *buf = ftrace_dyn_info_buffer;
4401 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4402 int r;
4403
4404 mutex_lock(&dyn_info_mutex);
4405 r = sprintf(buf, "%ld ", *p);
4406
4407 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4408 buf[r++] = '\n';
4409
4410 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4411
4412 mutex_unlock(&dyn_info_mutex);
4413
4414 return r;
4415}
4416
4417static const struct file_operations tracing_dyn_info_fops = {
4418 .open = tracing_open_generic,
4419 .read = tracing_read_dyn_info,
4420 .llseek = generic_file_llseek,
4421};
4422#endif
4423
4424static struct dentry *d_tracer;
4425
4426struct dentry *tracing_init_dentry(void)
4427{
4428 static int once;
4429
4430 if (d_tracer)
4431 return d_tracer;
4432
4433 if (!debugfs_initialized())
4434 return NULL;
4435
4436 d_tracer = debugfs_create_dir("tracing", NULL);
4437
4438 if (!d_tracer && !once) {
4439 once = 1;
4440 pr_warning("Could not create debugfs directory 'tracing'\n");
4441 return NULL;
4442 }
4443
4444 return d_tracer;
4445}
4446
4447static struct dentry *d_percpu;
4448
4449struct dentry *tracing_dentry_percpu(void)
4450{
4451 static int once;
4452 struct dentry *d_tracer;
4453
4454 if (d_percpu)
4455 return d_percpu;
4456
4457 d_tracer = tracing_init_dentry();
4458
4459 if (!d_tracer)
4460 return NULL;
4461
4462 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4463
4464 if (!d_percpu && !once) {
4465 once = 1;
4466 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4467 return NULL;
4468 }
4469
4470 return d_percpu;
4471}
4472
4473static void tracing_init_debugfs_percpu(long cpu)
4474{
4475 struct dentry *d_percpu = tracing_dentry_percpu();
4476 struct dentry *d_cpu;
4477 char cpu_dir[30]; /* 30 characters should be more than enough */
4478
4479 if (!d_percpu)
4480 return;
4481
4482 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4483 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4484 if (!d_cpu) {
4485 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4486 return;
4487 }
4488
4489 /* per cpu trace_pipe */
4490 trace_create_file("trace_pipe", 0444, d_cpu,
4491 (void *) cpu, &tracing_pipe_fops);
4492
4493 /* per cpu trace */
4494 trace_create_file("trace", 0644, d_cpu,
4495 (void *) cpu, &tracing_fops);
4496
4497 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4498 (void *) cpu, &tracing_buffers_fops);
4499
4500 trace_create_file("stats", 0444, d_cpu,
4501 (void *) cpu, &tracing_stats_fops);
4502
4503 trace_create_file("buffer_size_kb", 0444, d_cpu,
4504 (void *) cpu, &tracing_entries_fops);
4505}
4506
4507#ifdef CONFIG_FTRACE_SELFTEST
4508/* Let selftest have access to static functions in this file */
4509#include "trace_selftest.c"
4510#endif
4511
4512struct trace_option_dentry {
4513 struct tracer_opt *opt;
4514 struct tracer_flags *flags;
4515 struct dentry *entry;
4516};
4517
4518static ssize_t
4519trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4520 loff_t *ppos)
4521{
4522 struct trace_option_dentry *topt = filp->private_data;
4523 char *buf;
4524
4525 if (topt->flags->val & topt->opt->bit)
4526 buf = "1\n";
4527 else
4528 buf = "0\n";
4529
4530 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4531}
4532
4533static ssize_t
4534trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4535 loff_t *ppos)
4536{
4537 struct trace_option_dentry *topt = filp->private_data;
4538 unsigned long val;
4539 int ret;
4540
4541 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4542 if (ret)
4543 return ret;
4544
4545 if (val != 0 && val != 1)
4546 return -EINVAL;
4547
4548 if (!!(topt->flags->val & topt->opt->bit) != val) {
4549 mutex_lock(&trace_types_lock);
4550 ret = __set_tracer_option(current_trace, topt->flags,
4551 topt->opt, !val);
4552 mutex_unlock(&trace_types_lock);
4553 if (ret)
4554 return ret;
4555 }
4556
4557 *ppos += cnt;
4558
4559 return cnt;
4560}
4561
4562
4563static const struct file_operations trace_options_fops = {
4564 .open = tracing_open_generic,
4565 .read = trace_options_read,
4566 .write = trace_options_write,
4567 .llseek = generic_file_llseek,
4568};
4569
4570static ssize_t
4571trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4572 loff_t *ppos)
4573{
4574 long index = (long)filp->private_data;
4575 char *buf;
4576
4577 if (trace_flags & (1 << index))
4578 buf = "1\n";
4579 else
4580 buf = "0\n";
4581
4582 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4583}
4584
4585static ssize_t
4586trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4587 loff_t *ppos)
4588{
4589 long index = (long)filp->private_data;
4590 unsigned long val;
4591 int ret;
4592
4593 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4594 if (ret)
4595 return ret;
4596
4597 if (val != 0 && val != 1)
4598 return -EINVAL;
4599 set_tracer_flags(1 << index, val);
4600
4601 *ppos += cnt;
4602
4603 return cnt;
4604}
4605
4606static const struct file_operations trace_options_core_fops = {
4607 .open = tracing_open_generic,
4608 .read = trace_options_core_read,
4609 .write = trace_options_core_write,
4610 .llseek = generic_file_llseek,
4611};
4612
4613struct dentry *trace_create_file(const char *name,
4614 umode_t mode,
4615 struct dentry *parent,
4616 void *data,
4617 const struct file_operations *fops)
4618{
4619 struct dentry *ret;
4620
4621 ret = debugfs_create_file(name, mode, parent, data, fops);
4622 if (!ret)
4623 pr_warning("Could not create debugfs '%s' entry\n", name);
4624
4625 return ret;
4626}
4627
4628
4629static struct dentry *trace_options_init_dentry(void)
4630{
4631 struct dentry *d_tracer;
4632 static struct dentry *t_options;
4633
4634 if (t_options)
4635 return t_options;
4636
4637 d_tracer = tracing_init_dentry();
4638 if (!d_tracer)
4639 return NULL;
4640
4641 t_options = debugfs_create_dir("options", d_tracer);
4642 if (!t_options) {
4643 pr_warning("Could not create debugfs directory 'options'\n");
4644 return NULL;
4645 }
4646
4647 return t_options;
4648}
4649
4650static void
4651create_trace_option_file(struct trace_option_dentry *topt,
4652 struct tracer_flags *flags,
4653 struct tracer_opt *opt)
4654{
4655 struct dentry *t_options;
4656
4657 t_options = trace_options_init_dentry();
4658 if (!t_options)
4659 return;
4660
4661 topt->flags = flags;
4662 topt->opt = opt;
4663
4664 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4665 &trace_options_fops);
4666
4667}
4668
4669static struct trace_option_dentry *
4670create_trace_option_files(struct tracer *tracer)
4671{
4672 struct trace_option_dentry *topts;
4673 struct tracer_flags *flags;
4674 struct tracer_opt *opts;
4675 int cnt;
4676
4677 if (!tracer)
4678 return NULL;
4679
4680 flags = tracer->flags;
4681
4682 if (!flags || !flags->opts)
4683 return NULL;
4684
4685 opts = flags->opts;
4686
4687 for (cnt = 0; opts[cnt].name; cnt++)
4688 ;
4689
4690 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4691 if (!topts)
4692 return NULL;
4693
4694 for (cnt = 0; opts[cnt].name; cnt++)
4695 create_trace_option_file(&topts[cnt], flags,
4696 &opts[cnt]);
4697
4698 return topts;
4699}
4700
4701static void
4702destroy_trace_option_files(struct trace_option_dentry *topts)
4703{
4704 int cnt;
4705
4706 if (!topts)
4707 return;
4708
4709 for (cnt = 0; topts[cnt].opt; cnt++) {
4710 if (topts[cnt].entry)
4711 debugfs_remove(topts[cnt].entry);
4712 }
4713
4714 kfree(topts);
4715}
4716
4717static struct dentry *
4718create_trace_option_core_file(const char *option, long index)
4719{
4720 struct dentry *t_options;
4721
4722 t_options = trace_options_init_dentry();
4723 if (!t_options)
4724 return NULL;
4725
4726 return trace_create_file(option, 0644, t_options, (void *)index,
4727 &trace_options_core_fops);
4728}
4729
4730static __init void create_trace_options_dir(void)
4731{
4732 struct dentry *t_options;
4733 int i;
4734
4735 t_options = trace_options_init_dentry();
4736 if (!t_options)
4737 return;
4738
4739 for (i = 0; trace_options[i]; i++)
4740 create_trace_option_core_file(trace_options[i], i);
4741}
4742
4743static ssize_t
4744rb_simple_read(struct file *filp, char __user *ubuf,
4745 size_t cnt, loff_t *ppos)
4746{
4747 struct trace_array *tr = filp->private_data;
4748 struct ring_buffer *buffer = tr->buffer;
4749 char buf[64];
4750 int r;
4751
4752 if (buffer)
4753 r = ring_buffer_record_is_on(buffer);
4754 else
4755 r = 0;
4756
4757 r = sprintf(buf, "%d\n", r);
4758
4759 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4760}
4761
4762static ssize_t
4763rb_simple_write(struct file *filp, const char __user *ubuf,
4764 size_t cnt, loff_t *ppos)
4765{
4766 struct trace_array *tr = filp->private_data;
4767 struct ring_buffer *buffer = tr->buffer;
4768 unsigned long val;
4769 int ret;
4770
4771 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4772 if (ret)
4773 return ret;
4774
4775 if (buffer) {
4776 if (val)
4777 ring_buffer_record_on(buffer);
4778 else
4779 ring_buffer_record_off(buffer);
4780 }
4781
4782 (*ppos)++;
4783
4784 return cnt;
4785}
4786
4787static const struct file_operations rb_simple_fops = {
4788 .open = tracing_open_generic,
4789 .read = rb_simple_read,
4790 .write = rb_simple_write,
4791 .llseek = default_llseek,
4792};
4793
4794static __init int tracer_init_debugfs(void)
4795{
4796 struct dentry *d_tracer;
4797 int cpu;
4798
4799 trace_access_lock_init();
4800
4801 d_tracer = tracing_init_dentry();
4802
4803 trace_create_file("tracing_enabled", 0644, d_tracer,
4804 &global_trace, &tracing_ctrl_fops);
4805
4806 trace_create_file("trace_options", 0644, d_tracer,
4807 NULL, &tracing_iter_fops);
4808
4809 trace_create_file("tracing_cpumask", 0644, d_tracer,
4810 NULL, &tracing_cpumask_fops);
4811
4812 trace_create_file("trace", 0644, d_tracer,
4813 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4814
4815 trace_create_file("available_tracers", 0444, d_tracer,
4816 &global_trace, &show_traces_fops);
4817
4818 trace_create_file("current_tracer", 0644, d_tracer,
4819 &global_trace, &set_tracer_fops);
4820
4821#ifdef CONFIG_TRACER_MAX_TRACE
4822 trace_create_file("tracing_max_latency", 0644, d_tracer,
4823 &tracing_max_latency, &tracing_max_lat_fops);
4824#endif
4825
4826 trace_create_file("tracing_thresh", 0644, d_tracer,
4827 &tracing_thresh, &tracing_max_lat_fops);
4828
4829 trace_create_file("README", 0444, d_tracer,
4830 NULL, &tracing_readme_fops);
4831
4832 trace_create_file("trace_pipe", 0444, d_tracer,
4833 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4834
4835 trace_create_file("buffer_size_kb", 0644, d_tracer,
4836 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
4837
4838 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4839 &global_trace, &tracing_total_entries_fops);
4840
4841 trace_create_file("free_buffer", 0644, d_tracer,
4842 &global_trace, &tracing_free_buffer_fops);
4843
4844 trace_create_file("trace_marker", 0220, d_tracer,
4845 NULL, &tracing_mark_fops);
4846
4847 trace_create_file("saved_cmdlines", 0444, d_tracer,
4848 NULL, &tracing_saved_cmdlines_fops);
4849
4850 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4851 &trace_clock_fops);
4852
4853 trace_create_file("tracing_on", 0644, d_tracer,
4854 &global_trace, &rb_simple_fops);
4855
4856#ifdef CONFIG_DYNAMIC_FTRACE
4857 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4858 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4859#endif
4860
4861 create_trace_options_dir();
4862
4863 for_each_tracing_cpu(cpu)
4864 tracing_init_debugfs_percpu(cpu);
4865
4866 return 0;
4867}
4868
4869static int trace_panic_handler(struct notifier_block *this,
4870 unsigned long event, void *unused)
4871{
4872 if (ftrace_dump_on_oops)
4873 ftrace_dump(ftrace_dump_on_oops);
4874 return NOTIFY_OK;
4875}
4876
4877static struct notifier_block trace_panic_notifier = {
4878 .notifier_call = trace_panic_handler,
4879 .next = NULL,
4880 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4881};
4882
4883static int trace_die_handler(struct notifier_block *self,
4884 unsigned long val,
4885 void *data)
4886{
4887 switch (val) {
4888 case DIE_OOPS:
4889 if (ftrace_dump_on_oops)
4890 ftrace_dump(ftrace_dump_on_oops);
4891 break;
4892 default:
4893 break;
4894 }
4895 return NOTIFY_OK;
4896}
4897
4898static struct notifier_block trace_die_notifier = {
4899 .notifier_call = trace_die_handler,
4900 .priority = 200
4901};
4902
4903/*
4904 * printk is set to max of 1024, we really don't need it that big.
4905 * Nothing should be printing 1000 characters anyway.
4906 */
4907#define TRACE_MAX_PRINT 1000
4908
4909/*
4910 * Define here KERN_TRACE so that we have one place to modify
4911 * it if we decide to change what log level the ftrace dump
4912 * should be at.
4913 */
4914#define KERN_TRACE KERN_EMERG
4915
4916void
4917trace_printk_seq(struct trace_seq *s)
4918{
4919 /* Probably should print a warning here. */
4920 if (s->len >= 1000)
4921 s->len = 1000;
4922
4923 /* should be zero ended, but we are paranoid. */
4924 s->buffer[s->len] = 0;
4925
4926 printk(KERN_TRACE "%s", s->buffer);
4927
4928 trace_seq_init(s);
4929}
4930
4931void trace_init_global_iter(struct trace_iterator *iter)
4932{
4933 iter->tr = &global_trace;
4934 iter->trace = current_trace;
4935 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4936}
4937
4938static void
4939__ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4940{
4941 static arch_spinlock_t ftrace_dump_lock =
4942 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4943 /* use static because iter can be a bit big for the stack */
4944 static struct trace_iterator iter;
4945 unsigned int old_userobj;
4946 static int dump_ran;
4947 unsigned long flags;
4948 int cnt = 0, cpu;
4949
4950 /* only one dump */
4951 local_irq_save(flags);
4952 arch_spin_lock(&ftrace_dump_lock);
4953 if (dump_ran)
4954 goto out;
4955
4956 dump_ran = 1;
4957
4958 tracing_off();
4959
4960 /* Did function tracer already get disabled? */
4961 if (ftrace_is_dead()) {
4962 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4963 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4964 }
4965
4966 if (disable_tracing)
4967 ftrace_kill();
4968
4969 trace_init_global_iter(&iter);
4970
4971 for_each_tracing_cpu(cpu) {
4972 atomic_inc(&iter.tr->data[cpu]->disabled);
4973 }
4974
4975 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4976
4977 /* don't look at user memory in panic mode */
4978 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4979
4980 /* Simulate the iterator */
4981 iter.tr = &global_trace;
4982 iter.trace = current_trace;
4983
4984 switch (oops_dump_mode) {
4985 case DUMP_ALL:
4986 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4987 break;
4988 case DUMP_ORIG:
4989 iter.cpu_file = raw_smp_processor_id();
4990 break;
4991 case DUMP_NONE:
4992 goto out_enable;
4993 default:
4994 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4995 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4996 }
4997
4998 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4999
5000 /*
5001 * We need to stop all tracing on all CPUS to read the
5002 * the next buffer. This is a bit expensive, but is
5003 * not done often. We fill all what we can read,
5004 * and then release the locks again.
5005 */
5006
5007 while (!trace_empty(&iter)) {
5008
5009 if (!cnt)
5010 printk(KERN_TRACE "---------------------------------\n");
5011
5012 cnt++;
5013
5014 /* reset all but tr, trace, and overruns */
5015 memset(&iter.seq, 0,
5016 sizeof(struct trace_iterator) -
5017 offsetof(struct trace_iterator, seq));
5018 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5019 iter.pos = -1;
5020
5021 if (trace_find_next_entry_inc(&iter) != NULL) {
5022 int ret;
5023
5024 ret = print_trace_line(&iter);
5025 if (ret != TRACE_TYPE_NO_CONSUME)
5026 trace_consume(&iter);
5027 }
5028 touch_nmi_watchdog();
5029
5030 trace_printk_seq(&iter.seq);
5031 }
5032
5033 if (!cnt)
5034 printk(KERN_TRACE " (ftrace buffer empty)\n");
5035 else
5036 printk(KERN_TRACE "---------------------------------\n");
5037
5038 out_enable:
5039 /* Re-enable tracing if requested */
5040 if (!disable_tracing) {
5041 trace_flags |= old_userobj;
5042
5043 for_each_tracing_cpu(cpu) {
5044 atomic_dec(&iter.tr->data[cpu]->disabled);
5045 }
5046 tracing_on();
5047 }
5048
5049 out:
5050 arch_spin_unlock(&ftrace_dump_lock);
5051 local_irq_restore(flags);
5052}
5053
5054/* By default: disable tracing after the dump */
5055void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5056{
5057 __ftrace_dump(true, oops_dump_mode);
5058}
5059EXPORT_SYMBOL_GPL(ftrace_dump);
5060
5061__init static int tracer_alloc_buffers(void)
5062{
5063 int ring_buf_size;
5064 enum ring_buffer_flags rb_flags;
5065 int i;
5066 int ret = -ENOMEM;
5067
5068
5069 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5070 goto out;
5071
5072 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5073 goto out_free_buffer_mask;
5074
5075 /* Only allocate trace_printk buffers if a trace_printk exists */
5076 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5077 trace_printk_init_buffers();
5078
5079 /* To save memory, keep the ring buffer size to its minimum */
5080 if (ring_buffer_expanded)
5081 ring_buf_size = trace_buf_size;
5082 else
5083 ring_buf_size = 1;
5084
5085 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5086
5087 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5088 cpumask_copy(tracing_cpumask, cpu_all_mask);
5089
5090 /* TODO: make the number of buffers hot pluggable with CPUS */
5091 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5092 if (!global_trace.buffer) {
5093 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5094 WARN_ON(1);
5095 goto out_free_cpumask;
5096 }
5097 if (global_trace.buffer_disabled)
5098 tracing_off();
5099
5100
5101#ifdef CONFIG_TRACER_MAX_TRACE
5102 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5103 if (!max_tr.buffer) {
5104 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5105 WARN_ON(1);
5106 ring_buffer_free(global_trace.buffer);
5107 goto out_free_cpumask;
5108 }
5109#endif
5110
5111 /* Allocate the first page for all buffers */
5112 for_each_tracing_cpu(i) {
5113 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5114 max_tr.data[i] = &per_cpu(max_tr_data, i);
5115 }
5116
5117 set_buffer_entries(&global_trace,
5118 ring_buffer_size(global_trace.buffer, 0));
5119#ifdef CONFIG_TRACER_MAX_TRACE
5120 set_buffer_entries(&max_tr, 1);
5121#endif
5122
5123 trace_init_cmdlines();
5124
5125 register_tracer(&nop_trace);
5126 current_trace = &nop_trace;
5127 /* All seems OK, enable tracing */
5128 tracing_disabled = 0;
5129
5130 atomic_notifier_chain_register(&panic_notifier_list,
5131 &trace_panic_notifier);
5132
5133 register_die_notifier(&trace_die_notifier);
5134
5135 return 0;
5136
5137out_free_cpumask:
5138 free_cpumask_var(tracing_cpumask);
5139out_free_buffer_mask:
5140 free_cpumask_var(tracing_buffer_mask);
5141out:
5142 return ret;
5143}
5144
5145__init static int clear_boot_tracer(void)
5146{
5147 /*
5148 * The default tracer at boot buffer is an init section.
5149 * This function is called in lateinit. If we did not
5150 * find the boot tracer, then clear it out, to prevent
5151 * later registration from accessing the buffer that is
5152 * about to be freed.
5153 */
5154 if (!default_bootup_tracer)
5155 return 0;
5156
5157 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5158 default_bootup_tracer);
5159 default_bootup_tracer = NULL;
5160
5161 return 0;
5162}
5163
5164early_initcall(tracer_alloc_buffers);
5165fs_initcall(tracer_init_debugfs);
5166late_initcall(clear_boot_tracer);